The AMD 3rd Gen Ryzen Deep Dive Review: 3700X and 3900X Raising The Bar - AnandTech

It’s the review we’ve all been waiting for. Since December last year – and particularly since CES – AMD has been teasing us about the new Zen 2 microarchitecture and AMD’s newest Ryzen 3000 series of CPUs. Incorporating a significantly upgraded CPU architecture and built using TSMC's latest generation manufacturing process, AMD has continued to run at full speed at a time when rival Intel has struggled to move at all. The end result is that while the first and second generation of Ryzen CPUs were all about AMD returning to competition and eating into Intel's substantial performance lead, the Ryzen 3000 series is nothing less than AMD's first shot in nearly 13 years at meeting (or beating) Intel at their own game in the desktop CPU market. It's a big moment for AMD, and an exciting one in the CPU industry as a whole.

The new Ryzen 3000 chips mark the first "big" leap for AMD since they introduced their first Ryzen processors a bit over two years ago. Unlike last year's Ryzen 2000 series, which was a more minor refresh and brought some tweaks to the microarchitecture and process node, this year’s Ryzen 3000 is a major upgrade for both CPU architecture as well as on the manufacturing node. It marks AMD’s switch from GlobalFoundries' 12nm process to TSMC’s newest 7nm node. But what’s more exciting is how AMD was able to actually implement this switch: Ryzen 3000 isn’t merely a single chip, but a collection of non-uniform chiplets, introducing this design paradigm in a consumer product for the first time.

Today AMD launches its entire new CPU lineup and platform, alongside the new Navi-based Radeon RX 5700 series. In terms of CPU coverage, we’ll be taking a closer look at the new flagship, the $499 12-core Ryzen 3900X, as well as the $329 8-core Ryzen 7 3700X and its peculiar low TDP of 65W.

The CPU Line-up

AMD 'Matisse' Ryzen 3000 Series CPUs
AnandTech		Cores Threads		Base Freq	Boost Freq	L2 Cache	L3 Cache	PCIe 4.0	Chiplets IO+CPU	TDP	Price (SEP)
Ryzen 9	3950X	16C	32T	3.5	4.7	8 MB	64 MB	16+4+4	1+2	105W	$749
Ryzen 9	3900X	12C	24T	3.8	4.6	6 MB	64 MB	16+4+4	1+2	105W	$499
Ryzen 7	3800X	8C	16T	3.9	4.5	4 MB	32 MB	16+4+4	1+1	105W	$399
Ryzen 7	3700X	8C	16T	3.6	4.4	4 MB	32 MB	16+4+4	1+1	65W	$329
Ryzen 5	3600X	6C	12T	3.8	4.4	3 MB	32 MB	16+4+4	1+1	95W	$249
Ryzen 5	3600	6C	12T	3.6	4.2	3 MB	32 MB	16+4+4	1+1	65W	$199

AMD is launching 5 different SKUs today, with the 16-core Ryzen 9 3950X set to follow sometime in September. For today's launch AMD sampled the R9 3900X and R7 3700X, and we took them for a ride in the limited time we had with them, covering as much as we could.

Starting at the top we have the Ryzen 3900X, which is a 12-core design. In fact it's the first 12-core processor in a standard desktop socket, and it rather unique within AMD's product stack because it is currently the only SKU which takes full advantage of AMD’s newest chiplet architecture. Whereas all the other Ryzen parts are comprised of two chiplets – the base I/O die and a single CPU chiplet – 3900X comes with two such CPU chiplets, granting it (some of) the extra cores and the 64MB of L3 cache that entails.

Interestingly, while AMD has increased the core-count by 50% over its previous flagship processor, it has managed to keep the TDP to the same 105W as on the Ryzen 2700X. On top of this, the chip clocks in 300MHz faster than the predecessor in terms of boost clock, now reaching 4.6GHz; even the base clock has been increased by 100MHz, coming in at 3.8GHz. The big question then, is whether the new 7nm process node and Zen 2 are really this efficient, or should we be expecting more elevated power numbers?

Meanwhile our second chip of the day is the new Ryzen 3700X, which is configured and positioned as a particularly efficient model. With a boost clock of 4.4GHz and a base clock of 3.6GHz, the part should still be notably faster than the Ryzen 2700X, yet AMD has managed to make this a 65W TDP part which is going to make for some interesting analysis.

Continued Execution

Today’s Zen 2 and Ryzen 3000 launch is another step forward on AMD’s roadmap. The company has been working on a very ambitious development roadmap for their CPU designs, and Zen 2 is the company's first chance to flex their muscles and do a full iteration on their CPU core design.

Executing on this roadmap has been important for AMD both because it's helped them close the performance gap with Intel, and because it's helped to prove to customers (particularly hyperscalers and enterprises) that Zen wasn't a fluke, and that the company can reliably continue to improve its technology. This is especially noteworthy because while rival Intel hasn't been standing still, all of Intel's desktop technology for the last 3 years has been based on the same Skylake core architecture and variations on Intel's exceptional-for-the-time 14nm process. This will eventually change, as Intel gets their desktop house in order for 2020, but right now AMD is moving forward when Intel can not, allowing AMD to take full advantage of Intel's run of bad luck and wooing customers in the process.

And of course, AMD isn't done here. For the company’s engineers, today’s chips are last year’s work, and the company is working on the next generation Zen 3 core. Zen 3 is still a full generation away – and today is all about Zen 2 – but AMD is making it clear that Zen 2 isn't the end of the road either, and that they are working to further improve their CPU microarchitecture and overall platform.

Large Performance Boosts, Particularly for Gaming

The 3^rd gen Ryzen processors promise to bring some notably large performance improvements for users. The Zen 2 core microarchitecture is said to bring over 10% IPC improvements; this together with the higher clockspeeds should make for some solid generational improvements in a lot of workloads. For gaming in particular, AMD claims that we should be seeing some large improvements – the doubled size of the L3 cache is said to have made a notable mark on a lot of gaming titles, with AMD claiming ~20% to even ~30% improvements in some titles when compared to the last generation Ryzen 2700X.

Positioning the Ryzen 3000 series against Intel’s line-up is a matter of both performance as well as price. AMD had already made comparisons between the new SKUs and Intel’s counterparts back at Computex, where we saw comparisons between similarly priced units. According to the company, even Intel's pricey Skylake high-end desktop (HEDT) processor, the Core i9-9920X, isn't entirely out of the line of fire of the Ryzen 3900X.

As a quick recap to where things stand, compared to its immediate predecessor, Intel’s Coffee Lake Refresh received quite a bump in terms of both clock frequencies as well core counts. This allowed Intel to erase any Ryzen 2000 series lead in multi-threaded performance, all the while still maintaining a comfortable lead in single-threaded performance.

Against the Ryzen 3000 series, the Intel line-up will seemingly no longer have an IPC lead. AMD hasn’t been in such a situation since the Athlon 64 days well over 15 years ago, which is a remarkable turn of events. But with that said, make no mistake: IPC is just one half of the equation for single-threaded performance, and the other is raw frequency, and the Intel line-up still has a notable advantage thanks to its peak frequencies of up to 5GHz. So taking over Intel's single-threaded performance lead (at least on a consistent basis) is a tall order for the Ryzen 3000 series.

Comparison: Ryzen 9 3900X vs Core i9-9900K
AMD Ryzen 9 3900X	Features	Intel Core i9-9900K
12 / 24	Cores/Threads	8 / 16
3.8 / 4.6 GHz	Base/Turbo	3.6 / 5.0 GHz
16 (Free) + 4 (NVMe)	PCIe 4.0 Lanes	16 (Free) Gen 3.0 (No Gen 4.0)
512 KB/core	L2 Cache	256 KB/core
4x 16 MB 64 MB total	L3 Cache	16 MB
105 W	TDP	95 W
$499	Price (List)	$484

Taking a look at chip pricing and positioning then, the big flagship fight among desktop processors is going to be between the Ryzen 3900X at $484, and the i9-9900K at $484. Both of which happen to be the highest-end SKUs of their respective mainstream desktop computing platforms.

Here AMD should have a significant lead in terms of the multi-threaded performance of the new Ryzen 9 series as it’s able to employ 50% more cores than Intel, all while promising to remain in a similar TDP range of 105W vs 95W. We still expect the 9900K to win some workloads which are more lightly threaded simply due to Intel’s clock frequency lead, however this is something we’ll investigate more in detail in the coming benchmark analysis.

Comparison: Ryzen 7 3700X vs Core i7-9700K
AMD Ryzen 7 3700X	Features	Intel Core i7-9700K
8 / 16	Cores/Threads	8 / 8
3.6 / 4.4 GHz	Base/Turbo	3.6 / 4.9 GHz
16 (Free) + 4 (NVMe)	PCIe 4.0 Lanes	16 (Free) Gen 3.0 (No Gen 4.0)
512 KB/core	L2 Cache	256 KB/core
2x 16 MB 32MB total	L3 Cache	12 MB
65 W	TDP	95 W
$329	Price (List)	$385

The Ryzen 7 3700X is an interesting SKU. With only one populated CPU chiplet, the unit has half the available L3 cache versus the Ryzen 9 3900X. But it also has all the CPU cores within its one chiplet active. In theory this does mean that the CPU cores have less overall L3 cache available to them, as they have to share it with an additional core within their respective CCXs.

With a 3.6GHz/4.4Ghz base/boost clock configuration, we expect the 3700X to outperform the previous generation 2700X in all scenarios. The competition here based on pricing is the Core i7-9700K. Intel again should have a single-threaded performance advantage thanks to its 500 MHz higher clocks – but we’ll have to see how both chips match up in daily workloads.

We’ve already posted a microarchitecture overview and analysis of the Zen 2 microarchitecture following our Tech Day briefings in June, so be sure to read the piece in preparation for further testing analysis in our review today:

Read: AMD Zen 2 Microarchitecture Analysis: Ryzen 3000 and EPYC Rome

Among the biggest changes of the Ryzen 3000, alongside the improved core microarchitecture, is the chip’s overall cache hierarchy. The new chiplet houses CCXes with double the amount of L3, now 16MB instead of 8MB.

Furthermore the chiplet design with the introduction of the cIO die which houses the new memory controllers are undoubtedly going to have an impact on the memory latency and performance of the overall chip.

On the memory controller side particularly, AMD promises a wholly revamped design that brings new support for a whole lot faster DDR4 modules, with the chip coming by default categorized as supporting DDR4-3200, which is a bump over the DDR-2933 support of the Ryzen 2000 series.

AMD had published an interesting slide in regards to the new faster DDR support that went well above the officially supported 3200 speeds, with AMD claiming that the new controllers are able to support up to DDR4-4200 with ease and overclocking being possible to achieve ever higher speeds. However there’s a catch: in order to support DDR4 above 3600, the chip will automatically change the memory controller to infinity fabric clock ratio from being 1:1 to 2:1.

Whilst this doesn’t bottleneck the bandwidth of the memory to the cores as the new microarchitecture has now doubled the bus width of the Infinity Fabric to 512 bits, it does add a notable amount of cycles to the overall memory latency, meaning for the very vast majority of workloads, you’re better off staying at or under DDR4-3600 with a 1:1 MC:IF ratio. It’s to be noted that it’s still possible to maintain this 1:1 ratio by manually adjusting it at higher MC speeds, however stability of the system is no longer guaranteed as you’re effectively overclocking the Infinity Fabric as well in such a scenario.

For this article we didn’t have enough time to dive into the scaling behaviour of the different DRAM speeds, what we did investigate is a more architectural question of how exactly the new chiplet and cIO die architecture has impacted Zen2’s memory latency and memory performance.

To give better insights, we’re using my custom memory latency test that I use for mobile SoC testing and first covered in our review of the Galaxy S10+ and its two SoCs. Memory latency testing nowadays is a complicated topic as microarchitectures advance at a rapid rate, and in particular prefetchers can cause for sometimes misleading figures. Similarly, more brute-force approaches such as full random tests contain a lot of TLB miss latencies which don’t represent the actual structural latency of the system. Our custom latency suite thus isn’t a single one-number-fits-all test but rather a collection of tests that expose more details of the memory behaviour of the system.

The figures published on this page are run on DDR4-3200CL16 on the Ryzen 3900X and 2700X at timings of 16-16-16-36, and the i9-9900K was run with similar DDR4-3200CL16 at timings of 16-18-18-36.

  

Looking at the memory latency curves in a linear plotted graph, we see that there’s some larger obvious differences between the new Ryzen 3900X and the Ryzen 2700X. What immediately catches the eye when switching between the two results is the new 16MB L3 cache capacity which doubles upon the 8MB of Matisse. We have to remind ourselves that even though the whole chip contains 64MB of L3 cache, this is not a unified cache and a single CPU core will only see its own CCX’s L3 cache before going into main memory, which is in contrast to Intel’s L3 cache where all the cores have access to the full amount.

Before going into more details in the next graph, another thing that is obvious is that seemingly the 3900X’s DRAM latency is a tad worse than the 2700X’s. Among the many test patterns here the one to note is the “Structural Estimate” curve. This curve is actually a simple subtraction of the TLB+CLR Thrash tests minus the TLB Penalty figure. In the former, we’re causing as much cache-line replacement pressure as possible by repeatedly hitting the same cacheline within each memory page, also repeatedly trying to miss the TLB. In the latter, we’re still hitting the TLB heavily, but always using a different cache-line and thus having a minimum of cache-line pressure, resulting in an estimate of the TLB penalty. Subtracting the latter from the former gives us a quite good estimate of the actual structural latency of the chip and memory.

Now the big question is, why do it this way? I’ve found that with increasingly better prefetchers, it’s getting difficult in getting good memory latency numbers. Whilst it’s possible to just outright disable prefetchers on some platforms, that avenue isn’t always available.

Precisely when looking at the other various patterns in the graph, we’re seeing quite a large difference between the 3900X and the 2700X, with the 3900X showcasing notably lower latencies in a few of them. These figures are now a result of the new Zen2’s improved prefetchers which are able to better recognize patterns and pull out data from DRAM before the CPU core will handle that memory address.

  

Plotting the same data on a logarithmic graph, we better see some of the details.

In terms of the DRAM latency, it seems that the new Ryzen 3900X has regressed by around 10ns when compared to the 2700X (Note: Just take into the leading edge of the “Structural Estimate” figures as the better estimate) with ~74-75.5ns versus ~65.7ns.

It also looks like Zen2’s L3 cache has also gained a few cycles: A change from ~7.5ns at 4.3GHz to ~8.1ns at 4.6GHz would mean a regression from ~32 cycles to ~37 cycles. Such as change however was to be expected since doubling of the L3 cache structure has to come with some implementation compromises as there’s never just a free lunch. Zen2’s L3 cache latency is thus now about the same as Intel’s – while it was previously faster on Zen+.

Further interesting characteristics we see here is the increase of the capacity of the L2 TLB. This can be seen in the “TLB Penalty” curve, and the depth here corresponds to AMD’s published details of increasing the structure from 1536 pages to 2048 pages. It’s to be noted that the L3 capacity now exceeds the capacity of the TLB, meaning a single CPU core will have only the best access latencies to up to 8MB in the cache before starting to have to page-walk. A similar behaviour we see in the L2 cache where the L1 TLB capacity only covers 256KB of the cache before having to look up entries in the L2 TLB.

Another very interesting characteristic of AMD’s microarchitecture which contrasts Intel’s, is the fact that AMD prefetchers into the L2 cache, while Intel only does so for the nearest cache-line. Such a behaviour is a double-edged sword, on one hand AMD’s cores have can have better latencies to needed data, but on the other hand in the case of a unneeded prefetch, this put a lot more pressure on the L2 cache capacity, and could in effect counter-act some of the benefits of having double the capacity over Intel’s design.

  

Switching over to the memory bandwidth of the cache hierarchy, there’s one obvious new chance in the 3900X and Zen2: the inclusion of 256-bit wide datapaths. The new AGU and path changes mean that the core is able to now handle 256-bit AVX instruction once per cycle which is a doubling over the 128-bit datapaths of Zen and Zen+.

So while the bandwidth of 256-bit operations on the Ryzen 2700X looked identical to the 128-bit variants, the wider ops now on Zen2 effectively double the bandwidth of the core. This bandwidth doubling is evident in the L1 cache (The flip test is equal to a memory copy test), however the increase is only about 20% for the L2 and L3 caches.

There’s an interesting juxtaposition between AMD’s L3 cache bandwidth and Intel’s: AMD here has essentially as 60% advantage in bandwidth as the CCX’s L3 is much faster than Intel’s L3, when accessed by a single core. Particularly read-write modifications within a single cache-line (CLflip test) are significantly faster in both the L2 and L3 caches when compared to Intel’s core design.

Deeper into the DRAM regions, however we see that AMD is still lagging behind Intel when it comes to memory controller efficiency, so while the 3900X improves copy bandwidth from 19.2GB/s to 21GB/s, it still remains behind the 9900K’s 22.9GB/s. The store bandwidth (write bandwidth) to memory is also a tad lower on the AMD parts as the 3900X reaches 14.5GB/s versus Intel’s 18GB/s.

 

One aspect that AMD excels in is memory level parallelism. MLP is the ability for the CPU core to “park” memory accesses when they are missing the caches, and wait on them to return back later. In the above graph we see increasing number of random memory accesses depicted as the stacked lines, with the vertical axis showcasing the effective access speedup in relation to a single access.

Whilst both AMD and Intel’s MLP ability in the L2 are somewhat the same and reach 12 – this is because we’re saturating the bandwidth of the cache in this region and we just can’t go any faster via more accesses. In the L3 region however we see big differences between the two: While Intel starts off with around 20 accesses at the L3 with a 14-15x speedup, the TLBs and supporting core structures aren’t able to sustain this properly over the whole L3 as it’s having to access other L3 slices on the chip.

AMD’s implementation however seems to be able to handle over 32 accesses with an extremely robust 23x speedup. This advantage actually continues on to the DRAM region where we still see speed-ups up to 32 accesses, while Intel peaks at 16.

MLP ability is extremely important in order to be able to actually hide the various memory hierarchy latencies and to take full advantage of a CPU’s out-of-order execution abilities. AMD’s Zen cores here have seemingly the best microarchitecture in this regard, with only Apple’s mobile CPU cores having comparable characteristics. I think this was very much a conscious design choice of the microarchitecture as AMD knew their overall SoC design and future chiplet architecture would have to deal with higher latencies, and did their best in order to minimise such a disadvantage.

So while the new Zen2 cores do seemingly have worse off latencies, possibly a combined factor of a faster memory controller (faster frequencies could have come at a cost of latency in the implementation), a larger L3 but with additional cycles, it doesn’t mean that memory sensitive workloads will see much of a regression. AMD has been able to improve the core’s prefetchers, and average workload latency will be lower due to the doubled L3, and this is on top the core’s microarchitecture which seems to have outstandingly good MLP ability for whenever there is a cache miss, something to keep in mind as we investigate performance further.

Section by Gavin Bonshor

One of the biggest additions to AMD's AM4 socket is the introduction of the PCIe 4.0 interface. The new generation of X570 motherboards marks the first consumer motherboard chipset to feature PCIe 4.0 natively, which looks to offer users looking for even faster storage, and potentially better bandwidth for next-generation graphics cards over previous iterations of the current GPU architecture. We know that the Zen 2 processors have implemented the new TSMC 7nm manufacturing process with double the L3 cache compared with Zen 1. This new centrally focused IO chiplet is there regardless of the core count and uses the Infinity Fabric interconnect; the AMD X570 chipset uses four PCIe 4.0 lanes to uplink and downlink to the CPU IO die.

Looking at a direct comparison between AMD's AM4 X series chipsets, the X570 chipset adds PCIe 4.0 lanes over the previous X470 and X370's reliance on PCIe 3.0. A big plus point to the new X570 chipset is more support for USB 3.1 Gen2 with AMD allowing motherboard manufacturers to play with 12 flexible PCIe 4.0 lanes and implement features how they wish. This includes 8 x PCIe 4.0 lanes, with two blocks of PCIe 4.0 x4 to play with which vendors can add SATA, PCIe 4.0 x1 slots, and even support for 3 x PCIe 4.0 NVMe M.2 slots.

AMD X570, X470 and X370 Chipset Comparison
Feature	X570	X470	X370
PCIe Interface	4.0	3.0	3.0
Max PCH PCIe Lanes	24	24	24
USB 3.1 Gen2	8	2	2
Max USB 3.1 (Gen2/Gen1)	8/4	2/6	2/6
DDR4 Support	3200	2933	2667
Max SATA Ports	8	8	8
PCIe GPU Config	x16 x8/x8 x8/x8/x8*	x16 x8/x8 x8/x8/x4	x16 x8/x8 x8/x8/x4
Memory Channels (Dual)	2/2	2/2	2/2
Integrated 802.11ac WiFi MAC	N	N	N
Chipset TDP	15/11W**	6.8W	4.8W
Overclocking Support	Y	Y	Y
XFR2/PB2 Support	Y	Y	N

* Due to two different variations of the X570 chipset, one with a 15 W and another with an 11 W TDP, the extra power allows for more PCIe lanes, thus better GPU support overall. One example is the ASUS Pro WS X570-Ace model.
** Same reason as above, adding extra PCIe lanes to the chipset naturally increases power consumption.

One of the biggest changes in the chipset is within its architecture. The X570 chipset is the first chipset its manufactured in-house using ASMedia's IP, whereas previously with the X470 and X370 chipsets, ASMedia developed and produced it based on its 55nm architecture. While going from X370 at 6.8 W TDP at maximum load, X470 was improved upon in terms of power consumption to a lower TDP of 4.8 W. For X570, this has increased massively to an 11 W TDP on its consumer models, with a 15 W variant for its more professional and enterprise-focused models. The difference between the two X570 variations aside from power consumption is that the 15 W X570 chipset adds extra PCIe 4.0 lanes which seemingly increases power consumption greatly when compared to previous PCIe 3.0 focused chipsets.

Another major change due to the increased power consumption of the X570 chipset when compared to X470 and X370 is the cooling required. All but one of the launched product stack features an actively cooled chipset heatsink which is needed due to the increased power draw when using PCIe 4.0 due to the more complex implementation requirements over PCIe 3.0. While it is expected AMD will work on improving the TDP on future generations when using PCIe 4.0, it's forced manufacturers to implement more premium and more effective ways of keeping componentry on X570 cooler. This also stretches to the power delivery as AMD announced that a 16-core desktop Ryzen 3950X processor is set to launch later on in the year which means motherboard manufacturers need to implement better power deliveries, and better heatsinks capable of keeping the 105 W TDP processors efficient.

Memory support has also been improved with a seemingly better IMC on the Ryzen 3000 line-up when compared against the Ryzen 2000 and 1000 series of processors. Some motherboard vendors are advertising speeds of up to DDR4-4400 which until X570, was unheard of. X570 also marks a jump up to DDR4-3200 up from DDR4-2933 on X470, and DDR4-2667 on X370. As we investigated in our Ryzen 7 Memory Scaling piece back in 2017, we found out that the Infinity Fabric Interconnect scales well with frequency, and it is something that we will be analyzing once we get the launch of X570 out of the way, and potentially allow motherboard vendors to work on their infant firmware for AMD's new 7nm silicon.

Section by Dr. Ian Cutress (Orignal article)

One of the key points that have been a pain in the side of non-Intel processors using Windows has been the optimizations and scheduler arrangements in the operating system. We’ve seen in the past how Windows has not been kind to non-Intel microarchitecture layouts, such as AMD’s previous module design in Bulldozer, the Qualcomm hybrid CPU strategy with Windows on Snapdragon, and more recently with multi-die arrangements on Threadripper that introduce different memory latency domains into consumer computing.

Obviously AMD has a close relationship with Microsoft when it comes down to identifying a non-regular core topology with a processor, and the two companies work towards ensuring that thread and memory assignments, absent of program driven direction, attempt to make the most out of the system. With the May 10^th update to Windows, some additional features have been put in place to get the most out of the upcoming Zen 2 microarchitecture and Ryzen 3000 silicon layouts.

The optimizations come on two fronts, both of which are reasonably easy to explain.

Thread Grouping

The first is thread allocation. When a processor has different ‘groups’ of CPU cores, there are different ways in which threads are allocated, all of which have pros and cons. The two extremes for thread allocation come down to thread grouping and thread expansion.

Thread grouping is where as new threads are spawned, they will be allocated onto cores directly next to cores that already have threads. This keeps the threads close together, for thread-to-thread communication, however it can create regions of high power density, especially when there are many cores on the processor but only a couple are active.

Thread expansion is where cores are placed as far away from each other as possible. In AMD’s case, this would mean a second thread spawning on a different chiplet, or a different core complex/CCX, as far away as possible. This allows the CPU to maintain high performance by not having regions of high power density, typically providing the best turbo performance across multiple threads.

The danger of thread expansion is when a program spawns two threads that end up on different sides of the CPU. In Threadripper, this could even mean that the second thread was on a part of the CPU that had a long memory latency, causing an imbalance in the potential performance between the two threads, even though the cores those threads were on would have been at the higher turbo frequency.

Because of how modern software, and in particular video games, are now spawning multiple threads rather than relying on a single thread, and those threads need to talk to each other, AMD is moving from a hybrid thread expansion technique to a thread grouping technique. This means that one CCX will fill up with threads before another CCX is even accessed. AMD believes that despite the potential for high power density within a chiplet, while the other might be inactive, is still worth it for overall performance.

For Matisse, this should afford a nice improvement for limited thread scenarios, and on the face of the technology, gaming. It will be interesting to see how much of an affect this has on the upcoming EPYC Rome CPUs or future Threadripper designs. The single benchmark AMD provided in its explanation was Rocket League at 1080p Low, which reported a +15% frame rate gain.

Clock Ramping

For any of our users familiar with our Skylake microarchitecture deep dive, you may remember that Intel introduced a new feature called Speed Shift that enabled the processor to adjust between different P-states more freely, as well as ramping from idle to load very quickly – from 100 ms to 40ms in the first version in Skylake, then down to 15 ms with Kaby Lake. It did this by handing P-state control back from the OS to the processor, which reacted based on instruction throughput and request. With Zen 2, AMD is now enabling the same feature.

AMD already has sufficiently more granularity in its frequency adjustments over Intel, allowing for 25 MHz differences rather than 100 MHz differences, however enabling a faster ramp-to-load frequency jump is going to help AMD when it comes to very burst-driven workloads, such as WebXPRT (Intel’s favorite for this sort of demonstration). According to AMD, the way that this has been implemented with Zen 2 will require BIOS updates as well as moving to the Windows May 10^th update, but it will reduce frequency ramping from ~30 milliseconds on Zen to ~1-2 milliseconds on Zen 2. It should be noted that this is much faster than the numbers Intel tends to provide.

The technical name for AMD’s implementation involves CPPC2, or Collaborative Power Performance Control 2, and AMD’s metrics state that this can increase burst workloads and also application loading. AMD cites a +6% performance gain in application launch times using PCMark10’s app launch sub-test.

Hardened Security for Zen 2

Another aspect to Zen 2 is AMD’s approach to heightened security requirements of modern processors. As has been reported, a good number of the recent array of side channel exploits do not affect AMD processors, primarily because of how AMD manages its TLB buffers that have always required additional security checks before most of this became an issue. Nonetheless, for the issues to which AMD is vulnerable, it has implemented a full hardware-based security platform for them.

The change here comes for the Speculative Store Bypass, known as Spectre v4, which AMD now has additional hardware to work in conjunction with the OS or virtual memory managers such as hypervisors in order to control. AMD doesn’t expect any performance change from these updates. Newer issues such as Foreshadow and Zombieload do not affect AMD processors.

One big talking point around the new Ryzen 3000 series is the new augmented single-threaded performance of the new Zen 2 core. In order to investigate the topic in a more controlled manner with better documented workloads, we’ve fallen back to the industry standard SPEC benchmark suite.

We’ll be investigating the previous generation SPEC CPU2006 test suite giving us some better context to past platforms, as well as introducing the new SPEC CPU2017 suite. We have to note that SPEC2006 has been deprecated in favour of 2017, and we must also mention that the scores posted today are noted as estimates as they’re not officially submitted to the SPEC organisation.

For SPEC2006, we’re still using the same setup as on our mobile suite, meaning all the C/C++ benchmarks, while for SPEC2017 I’ve also went ahead and prepared all the Fortran tests for a near complete suite for desktop systems. I say near complete as due to time constraints we’re running the suite via WSL on Windows. I’ve checked that there are no noticeable performance differences to native Linux (we’re also compiling statically), however one bug on WSL is that it has a fixed stack size so we’ll be missing 521.wrf_r from the SPECfp2017 collection.

In terms of compilers, I’ve opted to use LLVM both for C/C++ and Fortran tests. For Fortran, we’re using the Flang compiler. The rationale of using LLVM over GCC is better cross-platform comparisons to platforms that have only have LLVM support and future articles where we’ll investigate this aspect more. We’re not considering closed-sourced compilers such as MSVC or ICC.

clang version 8.0.0-svn350067-1~exp1+0~20181226174230.701~1.gbp6019f2 (trunk)
clang version 7.0.1 (ssh://git@github.com/flang-compiler/flang-driver.git 
  24bd54da5c41af04838bbe7b68f830840d47fc03)

-Ofast -fomit-frame-pointer
-march=x86-64
-mtune=core-avx2 
-mfma -mavx -mavx2

Our compiler flags are straightforward, with basic –Ofast and relevant running and ISA switches to allow for AVX2 instructions.

The Ryzen 3900X system was run in the same way as the rest of our article with DDR4-3200CL16, same as with the i9-9900K, whilst the Ryzen 2700X had DDR-2933 with similar CL16 16-16-16-38 timings.

In terms of the int2006 benchmarks, the improvements of the new Zen2 based Ryzen 3900X is quite even across the board when compared to the Zen+ based Ryzen 2700X. We do note however somewhat larger performance increases in 403.gcc and 483.xalancbmk – it’s not immediately clear as to why as the benchmarks don’t have one particular characteristic that would fit Zen2’s design improvements, however I suspect it’s linked to the larger L3 cache.

445.gobmk in particular is a branch-heavy workload, and the 32% increase in performance here would be better explained by Zen2’s new additional TAGE branch predictor which is able to reduce overall branch misses.

It’s also interesting that although Ryzen3900X posted worse memory latency results than the 2700X, it’s still able to outperform the latter in memory sensitive workloads such as 429.mcf, although the increases for 471.omnetpp is amongst the smallest in the suite.

However we still see that AMD has an overall larger disadvantage to Intel in these memory sensitive tests, as the 9900K has large advantages in 429.mcf, and posting a large lead in the very memory bandwidth intensive 462.libquantum, the two tests that put the most pressure on the caches and memory subsystem.

In the fp2006 benchmarks, we gain see some larger jumps on the part of the Ryzen 3900X, particularly in 482.sphinx3. These two tests along with 450.soplex are characterized by higher data cache misses, so Zen2’s 16MB L3 cache should definitely be part of the reason we see such larger jumps.

I found it interesting that we’re not seeing much improvements in 470.lbm even though this is a test that is data store heavy, so I would have expected Zen2’s additional store AGU to greatly benefit this workload. There must be some higher level memory limitations which is bottlenecking the test.

453.povray isn’t data heavy nor branch heavy, as it’s one of the more simple workloads in the suite. Here it’s mostly up to the execution backend throughput and the ability of the front-end to feed it fast enough that are the bottlenecks. So while the Ryzen 3900X provides a big boost over the 2700X, it’s still largely lagging behind the 9900K, a characteristic we’re also seeing in the similar execution bottlenecked 456.hmmer of the integer suite.

Overall, the 3900X is 20.8% faster in the integer and floating point tests of the SPEC2006 suite, which corresponds to a 13% IPC increase, the metric that AMD officially uses to promote the Zen2 microarchitectural increases.

Moving on to the 2017 suite, we have to clarify that we’re using the Rate benchmark variations. The 2017 suite’s speed and rate benchmarks differ from each other in terms of workloads. The speed tests were designed for single-threaded testing and have large memory demands of up to 11GB, while the rate tests were meant for multi-process tests. We’re using the rate variations of the benchmarks because we don’t see any large differentiation between the two variations in terms of their characterisation and thus the performance scaling between the both should be extremely similar. On top of that, the rate benchmarks take up to 5x less time (+1 hour vs +6 hours), on top of being able to run them on more memory limited platforms which we plan on to do in the future.

In the int2017 suite, we’re seeing similar performance differences and improvements, although this time around there’s a few workloads that are a bit more limited in terms of their performance boosts on the new Ryzen 3900X.

Unfortunately I’m not quite as familiar with the exact characteristics of these tests as I am with the 2006 suite, so a more detailed analysis should follow in the next few months as we delve deeper into microarchitectural counters.

In the fp2017 suite, things are also quite even. Interesting enough here in particular AMD is able to leapfrog Intel’s 9900K in a lot more workloads, sometimes winning in terms of absolute performance and sometimes losing.

As for the overall performance scores, the new Ryzen 3900X improves by 18.1% over the 2700X. Although closing the gap greatly, it’s just shy of actually beating the 9900K’s absolute single-threaded performance.

Normalising the scores for frequency, we see that AMD has achieved something that the company hasn’t been able to claim in over 15 years: It has beat Intel in terms of overall IPC. Overall here, the IPC improvements over Zen+ are 10.5%, which is a bit lower than the 13% figure for SPEC2006.

We already know about Intel’s new upcoming Sunny Cove microarchitecture which should undoubtedly be able to regain the IPC crown with relative ease, but the question for Intel is if they’ll be able to still maintain the single-thread absolute performance crown and continue to see 5GHz or similar clock speeds with the new core design.

While more the focus of low-end and small form factor systems, web-based benchmarks are notoriously difficult to standardize. Modern web browsers are frequently updated, with no recourse to disable those updates, and as such there is difficulty in keeping a common platform. The fast paced nature of browser development means that version numbers (and performance) can change from week to week. Despite this, web tests are often a good measure of user experience: a lot of what most office work is today revolves around web applications, particularly email and office apps, but also interfaces and development environments. Our web tests include some of the industry standard tests, as well as a few popular but older tests.

We have also included our legacy benchmarks in this section, representing a stack of older code for popular benchmarks.

All of our benchmark results can also be found in our benchmark engine, Bench.

WebXPRT 3: Modern Real-World Web Tasks, including AI

The company behind the XPRT test suites, Principled Technologies, has recently released the latest web-test, and rather than attach a year to the name have just called it ‘3’. This latest test (as we started the suite) has built upon and developed the ethos of previous tests: user interaction, office compute, graph generation, list sorting, HTML5, image manipulation, and even goes as far as some AI testing.

For our benchmark, we run the standard test which goes through the benchmark list seven times and provides a final result. We run this standard test four times, and take an average.

Users can access the WebXPRT test at http://principledtechnologies.com/benchmarkxprt/webxprt/

WebXPRT 2015: HTML5 and Javascript Web UX Testing

The older version of WebXPRT is the 2015 edition, which focuses on a slightly different set of web technologies and frameworks that are in use today. This is still a relevant test, especially for users interacting with not-the-latest web applications in the market, of which there are a lot. Web framework development is often very quick but with high turnover, meaning that frameworks are quickly developed, built-upon, used, and then developers move on to the next, and adjusting an application to a new framework is a difficult arduous task, especially with rapid development cycles. This leaves a lot of applications as ‘fixed-in-time’, and relevant to user experience for many years.

Similar to WebXPRT3, the main benchmark is a sectional run repeated seven times, with a final score. We repeat the whole thing four times, and average those final scores.

Speedometer 2: JavaScript Frameworks

Our newest web test is Speedometer 2, which is a accrued test over a series of javascript frameworks to do three simple things: built a list, enable each item in the list, and remove the list. All the frameworks implement the same visual cues, but obviously apply them from different coding angles.

Our test goes through the list of frameworks, and produces a final score indicative of ‘rpm’, one of the benchmarks internal metrics. We report this final score.

Google Octane 2.0: Core Web Compute

A popular web test for several years, but now no longer being updated, is Octane, developed by Google. Version 2.0 of the test performs the best part of two-dozen compute related tasks, such as regular expressions, cryptography, ray tracing, emulation, and Navier-Stokes physics calculations.

The test gives each sub-test a score and produces a geometric mean of the set as a final result. We run the full benchmark four times, and average the final results.

Mozilla Kraken 1.1: Core Web Compute

Even older than Octane is Kraken, this time developed by Mozilla. This is an older test that does similar computational mechanics, such as audio processing or image filtering. Kraken seems to produce a highly variable result depending on the browser version, as it is a test that is keenly optimized for.

The main benchmark runs through each of the sub-tests ten times and produces an average time to completion for each loop, given in milliseconds. We run the full benchmark four times and take an average of the time taken.

Web Tests Analysis

Overall, in the web tests, the new Ryzen 3900X and 3700X perform very well with both chips showcasing quite large improvements over the 2700X.

We’re seeing quite an interesting match-up against Intel’s 9700K here which is leading the all the benchmarks. The reason for this is that SKU has SMT turned off. The singe-threaded performance advantage of this is that the CPU core no longer has to share the µOP cache structure between to different threads, and has the whole capacity dedicated to one thread. Web workloads in particular are amongst the most instruction pressure heavy workloads out there, and they benefit extremely from turning SMT off on modern cores.

Whilst we didn’t have the time yet to test the new 3900X and 3700X with SMT off, AMD’s core and op cache works the same in that it’s sharing the capacity amongst two threads, statically partitioning it. I’m pretty sure we’d see larger increases in the web benchmarks when turning off SMT as well, and we’ll be sure to revisit this particular point in the future.

Our System Test section focuses significantly on real-world testing, user experience, with a slight nod to throughput. In this section we cover application loading time, image processing, simple scientific physics, emulation, neural simulation, optimized compute, and 3D model development, with a combination of readily available and custom software. For some of these tests, the bigger suites such as PCMark do cover them (we publish those values in our office section), although multiple perspectives is always beneficial. In all our tests we will explain in-depth what is being tested, and how we are testing.

All of our benchmark results can also be found in our benchmark engine, Bench.

Application Load: GIMP 2.10.4

One of the most important aspects about user experience and workflow is how fast does a system respond. A good test of this is to see how long it takes for an application to load. Most applications these days, when on an SSD, load fairly instantly, however some office tools require asset pre-loading before being available. Most operating systems employ caching as well, so when certain software is loaded repeatedly (web browser, office tools), then can be initialized much quicker.

In our last suite, we tested how long it took to load a large PDF in Adobe Acrobat. Unfortunately this test was a nightmare to program for, and didn’t transfer over to Win10 RS3 easily. In the meantime we discovered an application that can automate this test, and we put it up against GIMP, a popular free open-source online photo editing tool, and the major alternative to Adobe Photoshop. We set it to load a large 50MB design template, and perform the load 10 times with 10 seconds in-between each. Due to caching, the first 3-5 results are often slower than the rest, and time to cache can be inconsistent, we take the average of the last five results to show CPU processing on cached loading.

Application loading is typically single thread limited, but we see here that at some point it also becomes core-resource limited. Having access to more resources per thread in a non-HT environment helps the 8C/8T and 6C/6T processors get ahead of both of the 5.0 GHz parts in our testing.

3D Particle Movement v2.1: Brownian Motion

Our 3DPM test is a custom built benchmark designed to simulate six different particle movement algorithms of points in a 3D space. The algorithms were developed as part of my PhD., and while ultimately perform best on a GPU, provide a good idea on how instruction streams are interpreted by different microarchitectures.

A key part of the algorithms is the random number generation – we use relatively fast generation which ends up implementing dependency chains in the code. The upgrade over the naïve first version of this code solved for false sharing in the caches, a major bottleneck. We are also looking at AVX2 and AVX512 versions of this benchmark for future reviews.

For this test, we run a stock particle set over the six algorithms for 20 seconds apiece, with 10 second pauses, and report the total rate of particle movement, in millions of operations (movements) per second. We have a non-AVX version and an AVX version, with the latter implementing AVX512 and AVX2 where possible.

3DPM v2.1 can be downloaded from our server: 3DPMv2.1.rar (13.0 MB)

With a non-AVX code base, the 9900K shows the IPC and frequency improvements over the R7 2700X, although in reality it is not as big of a percentage jump as you might imagine. The processors without HT get pushed back a bit here.

Dolphin 5.0: Console Emulation

One of the popular requested tests in our suite is to do with console emulation. Being able to pick up a game from an older system and run it as expected depends on the overhead of the emulator: it takes a significantly more powerful x86 system to be able to accurately emulate an older non-x86 console, especially if code for that console was made to abuse certain physical bugs in the hardware.

For our test, we use the popular Dolphin emulation software, and run a compute project through it to determine how close to a standard console system our processors can emulate. In this test, a Nintendo Wii would take around 1050 seconds.

The latest version of Dolphin can be downloaded from https://dolphin-emu.org/

DigiCortex 1.20: Sea Slug Brain Simulation

This benchmark was originally designed for simulation and visualization of neuron and synapse activity, as is commonly found in the brain. The software comes with a variety of benchmark modes, and we take the small benchmark which runs a 32k neuron / 1.8B synapse simulation, equivalent to a Sea Slug.

Example of a 2.1B neuron simulation

We report the results as the ability to simulate the data as a fraction of real-time, so anything above a ‘one’ is suitable for real-time work. Out of the two modes, a ‘non-firing’ mode which is DRAM heavy and a ‘firing’ mode which has CPU work, we choose the latter. Despite this, the benchmark is still affected by DRAM speed a fair amount.

DigiCortex can be downloaded from http://www.digicortex.net/

y-Cruncher v0.7.6: Microarchitecture Optimized Compute

I’ve known about y-Cruncher for a while, as a tool to help compute various mathematical constants, but it wasn’t until I began talking with its developer, Alex Yee, a researcher from NWU and now software optimization developer, that I realized that he has optimized the software like crazy to get the best performance. Naturally, any simulation that can take 20+ days can benefit from a 1% performance increase! Alex started y-cruncher as a high-school project, but it is now at a state where Alex is keeping it up to date to take advantage of the latest instruction sets before they are even made available in hardware.

For our test we run y-cruncher v0.7.6 through all the different optimized variants of the binary, single threaded and multi-threaded, including the AVX-512 optimized binaries. The test is to calculate 250m digits of Pi, and we use the single threaded and multi-threaded versions of this test.

Users can download y-cruncher from Alex’s website: http://www.numberworld.org/y-cruncher/

Agisoft Photoscan 1.3.3: 2D Image to 3D Model Conversion

One of the ISVs that we have worked with for a number of years is Agisoft, who develop software called PhotoScan that transforms a number of 2D images into a 3D model. This is an important tool in model development and archiving, and relies on a number of single threaded and multi-threaded algorithms to go from one side of the computation to the other.

In our test, we take v1.3.3 of the software with a good sized data set of 84 x 18 megapixel photos and push it through a reasonably fast variant of the algorithms, but is still more stringent than our 2017 test. We report the total time to complete the process.

Agisoft’s Photoscan website can be found here: http://www.agisoft.com/

Rendering is often a key target for processor workloads, lending itself to a professional environment. It comes in different formats as well, from 3D rendering through rasterization, such as games, or by ray tracing, and invokes the ability of the software to manage meshes, textures, collisions, aliasing, physics (in animations), and discarding unnecessary work. Most renderers offer CPU code paths, while a few use GPUs and select environments use FPGAs or dedicated ASICs. For big studios however, CPUs are still the hardware of choice.

All of our benchmark results can also be found in our benchmark engine, Bench.

Corona 1.3: Performance Render

An advanced performance based renderer for software such as 3ds Max and Cinema 4D, the Corona benchmark renders a generated scene as a standard under its 1.3 software version. Normally the GUI implementation of the benchmark shows the scene being built, and allows the user to upload the result as a ‘time to complete’.

We got in contact with the developer who gave us a command line version of the benchmark that does a direct output of results. Rather than reporting time, we report the average number of rays per second across six runs, as the performance scaling of a result per unit time is typically visually easier to understand.

The Corona benchmark website can be found at https://corona-renderer.com/benchmark

LuxMark v3.1: LuxRender via Different Code Paths

As stated at the top, there are many different ways to process rendering data: CPU, GPU, Accelerator, and others. On top of that, there are many frameworks and APIs in which to program, depending on how the software will be used. LuxMark, a benchmark developed using the LuxRender engine, offers several different scenes and APIs.

Taken from the Linux Version of LuxMark

In our test, we run the simple ‘Ball’ scene on both the C++ and OpenCL code paths, but in CPU mode. This scene starts with a rough render and slowly improves the quality over two minutes, giving a final result in what is essentially an average ‘kilorays per second’.

POV-Ray 3.7.1: Ray Tracing

The Persistence of Vision ray tracing engine is another well-known benchmarking tool, which was in a state of relative hibernation until AMD released its Zen processors, to which suddenly both Intel and AMD were submitting code to the main branch of the open source project. For our test, we use the built-in benchmark for all-cores, called from the command line.

POV-Ray can be downloaded from http://www.povray.org/

With the rise of streaming, vlogs, and video content as a whole, encoding and transcoding tests are becoming ever more important. Not only are more home users and gamers needing to convert video files into something more manageable, for streaming or archival purposes, but the servers that manage the output also manage around data and log files with compression and decompression. Our encoding tasks are focused around these important scenarios, with input from the community for the best implementation of real-world testing.

All of our benchmark results can also be found in our benchmark engine, Bench.

Handbrake 1.1.0: Streaming and Archival Video Transcoding

A popular open source tool, Handbrake is the anything-to-anything video conversion software that a number of people use as a reference point. The danger is always on version numbers and optimization, for example the latest versions of the software can take advantage of AVX-512 and OpenCL to accelerate certain types of transcoding and algorithms. The version we use here is a pure CPU play, with common transcoding variations.

We have split Handbrake up into several tests, using a Logitech C920 1080p60 native webcam recording (essentially a streamer recording), and convert them into two types of streaming formats and one for archival. The output settings used are:

• 720p60 at 6000 kbps constant bit rate, fast setting, high profile
• 1080p60 at 3500 kbps constant bit rate, faster setting, main profile
• 1080p60 HEVC at 3500 kbps variable bit rate, fast setting, main profile

7-zip v1805: Popular Open-Source Encoding Engine

Out of our compression/decompression tool tests, 7-zip is the most requested and comes with a built-in benchmark. For our test suite, we’ve pulled the latest version of the software and we run the benchmark from the command line, reporting the compression, decompression, and a combined score.

It is noted in this benchmark that the latest multi-die processors have very bi-modal performance between compression and decompression, performing well in one and badly in the other. There are also discussions around how the Windows Scheduler is implementing every thread. As we get more results, it will be interesting to see how this plays out.

Please note, if you plan to share out the Compression graph, please include the Decompression one. Otherwise you’re only presenting half a picture.

WinRAR 5.60b3: Archiving Tool

My compression tool of choice is often WinRAR, having been one of the first tools a number of my generation used over two decades ago. The interface has not changed much, although the integration with Windows right click commands is always a plus. It has no in-built test, so we run a compression over a set directory containing over thirty 60-second video files and 2000 small web-based files at a normal compression rate.

WinRAR is variable threaded but also susceptible to caching, so in our test we run it 10 times and take the average of the last five, leaving the test purely for raw CPU compute performance.

AES Encryption: File Security

A number of platforms, particularly mobile devices, are now offering encryption by default with file systems in order to protect the contents. Windows based devices have these options as well, often applied by BitLocker or third-party software. In our AES encryption test, we used the discontinued TrueCrypt for its built-in benchmark, which tests several encryption algorithms directly in memory.

The data we take for this test is the combined AES encrypt/decrypt performance, measured in gigabytes per second. The software does use AES commands for processors that offer hardware selection, however not AVX-512.

The Office test suite is designed to focus around more industry standard tests that focus on office workflows, system meetings, some synthetics, but we also bundle compiler performance in with this section. For users that have to evaluate hardware in general, these are usually the benchmarks that most consider.

All of our benchmark results can also be found in our benchmark engine, Bench.

PCMark 10: Industry Standard System Profiler

Futuremark, now known as UL, has developed benchmarks that have become industry standards for around two decades. The latest complete system test suite is PCMark 10, upgrading over PCMark 8 with updated tests and more OpenCL invested into use cases such as video streaming.

PCMark splits its scores into about 14 different areas, including application startup, web, spreadsheets, photo editing, rendering, video conferencing, and physics. We post all of these numbers in our benchmark database, Bench, however the key metric for the review is the overall score.

We're investigating the PCMark results, which seem abnormally high.

3DMark Physics: In-Game Physics Compute

Alongside PCMark is 3DMark, Futuremark’s (UL’s) gaming test suite. Each gaming tests consists of one or two GPU heavy scenes, along with a physics test that is indicative of when the test was written and the platform it is aimed at. The main overriding tests, in order of complexity, are Ice Storm, Cloud Gate, Sky Diver, Fire Strike, and Time Spy.

Some of the subtests offer variants, such as Ice Storm Unlimited, which is aimed at mobile platforms with an off-screen rendering, or Fire Strike Ultra which is aimed at high-end 4K systems with lots of the added features turned on. Time Spy also currently has an AVX-512 mode (which we may be using in the future).

For our tests, we report in Bench the results from every physics test, but for the sake of the review we keep it to the most demanding of each scene: Ice Storm Unlimited, Cloud Gate, Sky Diver, Fire Strike Ultra, and Time Spy.

The older Ice Storm test didn't much like the Core i9-9900K, pushing it back behind the R7 1800X. For the more modern tests focused on PCs, the 9900K wins out. The lack of HT is hurting the other two parts.

GeekBench4: Synthetics

A common tool for cross-platform testing between mobile, PC, and Mac, GeekBench 4 is an ultimate exercise in synthetic testing across a range of algorithms looking for peak throughput. Tests include encryption, compression, fast Fourier transform, memory operations, n-body physics, matrix operations, histogram manipulation, and HTML parsing.

I’m including this test due to popular demand, although the results do come across as overly synthetic, and a lot of users often put a lot of weight behind the test due to the fact that it is compiled across different platforms (although with different compilers).

We record the main subtest scores (Crypto, Integer, Floating Point, Memory) in our benchmark database, but for the review we post the overall single and multi-threaded results.

3DPM v1: Naïve Code Variant of 3DPM v2.1

The first legacy test in the suite is the first version of our 3DPM benchmark. This is the ultimate naïve version of the code, as if it was written by scientist with no knowledge of how computer hardware, compilers, or optimization works (which in fact, it was at the start). This represents a large body of scientific simulation out in the wild, where getting the answer is more important than it being fast (getting a result in 4 days is acceptable if it’s correct, rather than sending someone away for a year to learn to code and getting the result in 5 minutes).

In this version, the only real optimization was in the compiler flags (-O2, -fp:fast), compiling it in release mode, and enabling OpenMP in the main compute loops. The loops were not configured for function size, and one of the key slowdowns is false sharing in the cache. It also has long dependency chains based on the random number generation, which leads to relatively poor performance on specific compute microarchitectures.

3DPM v1 can be downloaded with our 3DPM v2 code here: 3DPMv2.1.rar (13.0 MB)

x264 HD 3.0: Older Transcode Test

This transcoding test is super old, and was used by Anand back in the day of Pentium 4 and Athlon II processors. Here a standardized 720p video is transcoded with a two-pass conversion, with the benchmark showing the frames-per-second of each pass. This benchmark is single-threaded, and between some micro-architectures we seem to actually hit an instructions-per-clock wall.

CineBench 11.5 and 10

Cinebench is a widely known benchmarking tool for measuring performance relative to MAXON's animation software Cinema 4D. Cinebench has been optimized over a decade and focuses on purely CPU horsepower, meaning if there is a discrepancy in pure throughput characteristics, Cinebench is likely to show that discrepancy. Arguably other software doesn't make use of all the tools available, so the real world relevance might purely be academic, but given our large database of data for Cinebench it seems difficult to ignore a small five minute test. We run the modern version 15 in this test, as well as the older 11.5 due to our back data.

Albeit different to most of the other commonly played MMO or massively multiplayer online games, World of Tanks is set in the mid-20th century and allows players to take control of a range of military based armored vehicles. World of Tanks (WoT) is developed and published by Wargaming who are based in Belarus, with the game’s soundtrack being primarily composed by Belarusian composer Sergey Khmelevsky. The game offers multiple entry points including a free-to-play element as well as allowing players to pay a fee to open up more features. One of the most interesting things about this tank based MMO is that it achieved eSports status when it debuted at the World Cyber Games back in 2012.

World of Tanks enCore is a demo application for a new and unreleased graphics engine penned by the Wargaming development team. Over time the new core engine will implemented into the full game upgrading the games visuals with key elements such as improved water, flora, shadows, lighting as well as other objects such as buildings. The World of Tanks enCore demo app not only offers up insight into the impending game engine changes, but allows users to check system performance to see if the new engine run optimally on their system.

AnandTech CPU Gaming 2019 Game List
Game	Genre	Release Date	API	IGP	Low	Med	High
World of Tanks enCore	Driving / Action	Feb 2018	DX11	768p Minimum	1080p Medium	1080p Ultra	4K Ultra

All of our benchmark results can also be found in our benchmark engine, Bench.

World of Tanks enCore	IGP	Low	Medium	High
Average FPS
95th Percentile

Next up is Middle-earth: Shadow of War, the sequel to Shadow of Mordor. Developed by Monolith, whose last hit was arguably F.E.A.R., Shadow of Mordor returned them to the spotlight with an innovative NPC rival generation and interaction system called the Nemesis System, along with a storyline based on J.R.R. Tolkien's legendarium, and making it work on a highly modified engine that originally powered F.E.A.R. in 2005.

Using the new LithTech Firebird engine, Shadow of War improves on the detail and complexity, and with free add-on high-resolution texture packs, offers itself as a good example of getting the most graphics out of an engine that may not be bleeding edge. Shadow of War also supports HDR (HDR10).

AnandTech CPU Gaming 2019 Game List
Game	Genre	Release	API	IGP	Low	Med	High
Shadow of War	Action / RPG	Sep 2017	DX11	720p Ultra	1080p Ultra	4K High	8K High

All of our benchmark results can also be found in our benchmark engine, Bench.

Shadow of War	IGP	Low	Medium	High
Average FPS

Seen as the holy child of DirectX12, Ashes of the Singularity (AoTS, or just Ashes) has been the first title to actively go explore as many of the DirectX12 features as it possibly can. Stardock, the developer behind the Nitrous engine which powers the game, has ensured that the real-time strategy title takes advantage of multiple cores and multiple graphics cards, in as many configurations as possible.

As a real-time strategy title, Ashes is all about responsiveness during both wide open shots but also concentrated battles. With DirectX12 at the helm, the ability to implement more draw calls per second allows the engine to work with substantial unit depth and effects that other RTS titles had to rely on combined draw calls to achieve, making some combined unit structures ultimately very rigid.

Stardock clearly understand the importance of an in-game benchmark, ensuring that such a tool was available and capable from day one, especially with all the additional DX12 features used and being able to characterize how they affected the title for the developer was important. The in-game benchmark performs a four minute fixed seed battle environment with a variety of shots, and outputs a vast amount of data to analyze.

For our benchmark, we run Ashes Classic: an older version of the game before the Escalation update. The reason for this is that this is easier to automate, without a splash screen, but still has a strong visual fidelity to test.

AnandTech CPU Gaming 2019 Game List
Game	Genre	Release Date	API	IGP	Low	Med	High
Ashes: Classic	RTS	Mar 2016	DX12	720p Standard	1080p Standard	1440p Standard	4K Standard

Ashes has dropdown options for MSAA, Light Quality, Object Quality, Shading Samples, Shadow Quality, Textures, and separate options for the terrain. There are several presents, from Very Low to Extreme: we run our benchmarks at the above settings, and take the frame-time output for our average and percentile numbers.

All of our benchmark results can also be found in our benchmark engine, Bench.

Ashes Classic	IGP	Low	Medium	High
Average FPS
95th Percentile

Strange Brigade is based in 1903’s Egypt and follows a story which is very similar to that of the Mummy film franchise. This particular third-person shooter is developed by Rebellion Developments which is more widely known for games such as the Sniper Elite and Alien vs Predator series. The game follows the hunt for Seteki the Witch Queen who has arose once again and the only ‘troop’ who can ultimately stop her. Gameplay is cooperative centric with a wide variety of different levels and many puzzles which need solving by the British colonial Secret Service agents sent to put an end to her reign of barbaric and brutality.

The game supports both the DirectX 12 and Vulkan APIs and houses its own built-in benchmark which offers various options up for customization including textures, anti-aliasing, reflections, draw distance and even allows users to enable or disable motion blur, ambient occlusion and tessellation among others. AMD has boasted previously that Strange Brigade is part of its Vulkan API implementation offering scalability for AMD multi-graphics card configurations.

AnandTech CPU Gaming 2019 Game List
Game	Genre	Release Date	API	IGP	Low	Med	High
Strange Brigade*	FPS	Aug 2018	DX12 Vulkan	720p Low	1080p Medium	1440p High	4K Ultra
*Strange Brigade is run in DX12 and Vulkan modes

All of our benchmark results can also be found in our benchmark engine, Bench.

Strange Brigade DX12	IGP	Low	Medium	High
Average FPS
95th Percentile

Strange Brigade Vulkan	IGP	Low	Medium	High
Average FPS
95th Percentile

The highly anticipated iteration of the Grand Theft Auto franchise hit the shelves on April 14th 2015, with both AMD and NVIDIA in tow to help optimize the title. GTA doesn’t provide graphical presets, but opens up the options to users and extends the boundaries by pushing even the hardest systems to the limit using Rockstar’s Advanced Game Engine under DirectX 11. Whether the user is flying high in the mountains with long draw distances or dealing with assorted trash in the city, when cranked up to maximum it creates stunning visuals but hard work for both the CPU and the GPU.

For our test we have scripted a version of the in-game benchmark. The in-game benchmark consists of five scenarios: four short panning shots with varying lighting and weather effects, and a fifth action sequence that lasts around 90 seconds. We use only the final part of the benchmark, which combines a flight scene in a jet followed by an inner city drive-by through several intersections followed by ramming a tanker that explodes, causing other cars to explode as well. This is a mix of distance rendering followed by a detailed near-rendering action sequence, and the title thankfully spits out frame time data.

AnandTech CPU Gaming 2019 Game List
Game	Genre	Release Date	API	IGP	Low	Med	High
Grand Theft Auto V	Open World	Apr 2015	DX11	720p Low	1080p High	1440p Very High	4K Ultra
*Strange Brigade is run in DX12 and Vulkan modes

There are no presets for the graphics options on GTA, allowing the user to adjust options such as population density and distance scaling on sliders, but others such as texture/shadow/shader/water quality from Low to Very High. Other options include MSAA, soft shadows, post effects, shadow resolution and extended draw distance options. There is a handy option at the top which shows how much video memory the options are expected to consume, with obvious repercussions if a user requests more video memory than is present on the card (although there’s no obvious indication if you have a low end GPU with lots of GPU memory, like an R7 240 4GB).

All of our benchmark results can also be found in our benchmark engine, Bench.

GTA 5	IGP	Low	Medium	High
Average FPS
95th Percentile

Aside from keeping up-to-date on the Formula One world, F1 2017 added HDR support, which F1 2018 has maintained; otherwise, we should see any newer versions of Codemasters' EGO engine find its way into F1. Graphically demanding in its own right, F1 2018 keeps a useful racing-type graphics workload in our benchmarks.

Aside from keeping up-to-date on the Formula One world, F1 2017 added HDR support, which F1 2018 has maintained. We use the in-game benchmark, set to run on the Montreal track in the wet, driving as Lewis Hamilton from last place on the grid. Data is taken over a one-lap race.

AnandTech CPU Gaming 2019 Game List
Game	Genre	Release Date	API	IGP	Low	Med	High
F1 2018	Racing	Aug 2018	DX11	720p Low	1080p Med	4K High	4K Ultra

All of our benchmark results can also be found in our benchmark engine, Bench.

F1 2018	IGP	Low	Medium	High
Average FPS
95th Percentile

Power consumption of the new Ryzen 3900 and 3700X are of particular interest because it’s a very key aspect of the new generation chipsets, and AMD promises some extremely large improvements thanks to the new 7nm process node as well as the optimised chiplet design.

When comparing the single-chiplet Ryzen 3700X to the previous generation Ryzen 2700X, we’re seeing quite some dramatic differences in core power consumption. In particular power consumption at each chip’s respective peak frequency is notably different: Although the new 3700X has a 100MHz higher clock speed and thus is further up the exponential power curve, it manages to showcase 32% lower absolute power than the 2700X.

We have to remember that we’re talking about overall absolute power, and not efficiency of the chip. When taking actual performance into account through the higher clock as well as Zen2’s increased performance per clock, the Performance/W figures for the new 3700X should be significantly higher than its predecessor.

What is curious about the new chip is just how closely it follows its power limitations. The new boosting algorithm on the Ryzen 3 series is a particularly “opportunistic” one that will go as high in frequencies as it can go within its constraints, no matter the amount of CPU cores.

The constraints are as follows:

• Package Power Tracking (PPT): The power threshold that is allowed to be delivered to the socket.
  • This is 88W for 65W TDP processors, and 142W for 105W TDP processors.
• Thermal Design Current (TDC): The maximum amount of current delivered by the motherboard’s voltage regulators when under thermally constrained scenarios (high temperatures)
  • This is 60A for 65W TDP processors, and 95A for 105W TDP processors.
• Electrical Design Current (EDC): This is the maximum amount of current at any instantaneous short period of time that can be delivered by the motherboard’s voltage regulators.
  • This is 90A for 65W TDP processors, and 140A for 105W TDP processors.

Looking at the total power consumption of the new 3700X, the chip is very much seemingly hitting and maintaining the 88W PPT limitations of the default settings, and we’re measuring 90W peak consumption across the package.

When having a closer look at the new Ryzen 9 3900X, first we have to enjoy the sheer amount of cores of this processor!

Following that, we see that this CPU’s per-core peak power consumption is quite notably higher than that of the 3700X, which is not a surprise given that the chip is clocked 200MHz higher at 4.6GHz versus “just” 4.4GHz. However even at this much higher clock, the 3900X’s power consumption remains notably lower than that of the 2700X.

Scaling up in threads as well as cores, we’re seeing a similar scaling behaviour, with the large difference being that the 3900X is maintaining higher power consumption per core (and frequency) than the 3700X. Fully loading the chip we’re seeing 118W power on the CPU cores while the package power is falling in at the exact 142W that AMD describes as the PPT limit of 105W TDP processors such as the 3900X.

Another thing to note in the results between the 3700X results and the 3900X, is that un-core power on the latter is quite higher. This really shouldn’t come as a surprise as the processor has a second chiplet who will have L3 and Infinity Fabric that will use more power.

Graphing the three processors together, we see two main aspects: Again the 3900X and 3700X both consuming notably less power than the 2700X, and the 3700X’s hard limit when reaching the 88W PPT limit while the 3900X is able to scale further up till it hits the 142W limit.

Comparing the full load power characteristics of both SKUs, they end up extremely competitive in both their respective categories. The 3700X’s 90W hard-limit puts it at the very bottom of the CPUs we’ve used in our testing today, which is quite astonishing as the chip is trading blows with the 9700K and 9900K across all of our test workloads, and the latter chip’s power consumption is well over 60% above the 3700X’s.

The 3900X is also impressive given that it’s a 12-core CPU. While posting substantial performance improvements of the 12-core Threadripper counterparts, the 3900X still manages to be significantly less thermally constrained thanks to its much lower power consumption, peaking in at 142W.

The most interesting aspect of AMD’s new opportunistic power boost mechanism lies in a CPU we weren’t able to test today: the Ryzen 7 3800X. At stock behaviour, the chip’s 105W TDP should allow it to behave a lot more like the 3900X when it comes to the higher thread-count frequencies, at least until it maxes out its 8 cores on its single chiplet, which might really put it ahead of the 3700X in terms of multi-threaded performance workloads.

Overclocking: PBO & All-Core

In POV-Ray, running the 3900X at a flat 4.3GHz gives it a 8.2% performance boost over stock. Enabling PBO doesn’t make much difference in multi-threaded workloads for the 3900X as it’s still being limited by the 142W PPT limit.

Unfortunately we weren’t able to further investigate raising the PPT limit for this article due to time contraints as well as currently non-final firmware version for X570 motherboards from the vendors.

Turning on PBO will increase the single-threaded performance of the 3900X by a few percent, scoring just slightly higher than the stock settings. Naturally the 4.3 GHz flat overclock will regress in performance as it loses out 300MHz peak frequency compared to stock.

Finally, a Cinebench R15 MT run shows similar multi-threaded behaviour, with the 4.3GHz flat overclock achieving a 9.2% better score, whilst the PBO overclock isn’t able to further increase frequencies beyond the default power limits of the chip.

Overall, we’ve been eagerly awaiting today’s launch for months, and all the while AMD has certainly given us some high expectations for their 3^rd generation Ryzen CPUs. At the end of the day I think that AMD was able to deliver on all of their promises, and hitting all of the performance targets that they needed to. Furthermore, where AMD kills it is in terms of value, as both the 3700X and the 3900X really deliver in terms of offering outstanding alternatives to the competition.

The New Zen 2 µarch & Chiplet Design

The basis for the new 3^rd generation Ryzen processors is AMD’s new high-risk high-reward bet on moving away from a single monolithic die to a chiplet-based MCM (Multi-chip module) design. What this has allowed AMD to do is to maximise the performance characteristics of their 7nm design for the new Ryzen 3000 chipsets. Meanwhile, having the I/O components and the memory controllers on a 12nm process node not only allows AMD to minimise the cost of the platform, but also allows them to optimise the silicon for their specific use-cases.

The actual CPU chiplets (CPU-lets?) are manufactured on TSMC’s leading edge 7nm process node and AMD has seemingly been able to take full advantage of the process, not only lowering the power consumption of the cores, but actually also raising the clock frequency at the same time, bringing for some impressive power efficiency benefits.

The new design did seemingly make some compromises, and we saw that the DRAM memory latency of this new system architecture is slower than the previous monolithic implementation. However, here is also where things get interesting. Even though this is a theoretical regression on paper, when it comes to actual performance in workloads the regression is essentially non-existent, and AMD is able to showcase improvements even in the most memory-sensitive workloads. This is thanks to the new Zen 2 CPU core’s improved microarchitecture, with new improved prefetchers and overall outstanding Memory Level Parallelism (MLP) designs. Further helping AMD's memory/cache situation is the doubling of the CCX’s L3 cache from 8MB to 16MB, which on average, ends up with better workload memory performance.

Not that Zen 2 is soley about memory performance, either. The CPU core's front-end improvements such as the new TAGE predictor – and in particular the much increased capacity of the operation cache – is very visible in some workloads. We’ve also seen the core’s new 256-bit (AVX2) vector datapaths to work very well.

In the majority of controlled tests, AMD has done something they haven’t been able to achieve in almost 15 years, since the tail-end of the Athlon 64's reign in 2005: that is to have a CPU microarchitecture with higher performance per clock than Intel's leading architecture. Zen 2 finally achieves this symbolic mark by a hair’s margin, with the new core improving IPC by 10-13% when compared to Zen+.

Having said that, Intel still very much holds the single-threaded performance crown by a few percent. Intel’s higher achieved frequencies as well as continued larger lead in memory sensitive workloads are still goals that AMD has to work towards to, and future Zen iterations will have to further improve in order to have a shot at the ST performance crown.

Beyond this, it’s remarkable that AMD has been able to achieve all of this while having significantly lower power consumption than Intel's best desktop chip, all thanks to the new process node.

The 3700X & 3900X Versus The Competition, Verdict

Office CPU Performance and Productivity

It’s in these categories where AMD’s strengths lie: In the majority of our systems benchmarks, AMD more often than not is able to lead Intel’s 9700K and 9900K in terms of performance. Particularly it was interesting to see the new 3^rd gen Ryzens post larger improvements in the web tests, all thanks to Zen 2’s improved and larger op cache.

In anything that is remotely multi-threaded, AMD is also able to take the performance crown, with only Intel’s HEDT i9-7920X being able to top the new 12-core Ryzen 3900X. The 3700X here still hangs in there being extremely competitive, falling in-between the 9700K and 9900K when it comes to multi-threaded workloads, sometimes even beating the 9900K in some workloads, a respectable result.

Gaming Performance

When it comes to gaming performance, the 9700K and 9900K remain the best performing CPUs on the market.

That being said, the new 3700X and 3900X are posting enormous improvements over the 2700X, and we can confirm AMD’s claims of up to 30-35% better performance in some games over the 2700X.

Here’s the thing: while AMD does still lag behind Intel in gaming performance, the gap has narrowed immensely to the point that the Ryzen CPUs are no longer something to be dismissed if you want to have a high-end gaming machine, and are still very much a viable option worth considering.

Everything Tied Together: A Win For AMD

What really does make the Ryzen 3700X and 3900X winners in my eyes is their overall packages and performance. They’re outstanding all-rounders and AMD has managed to vastly improve some of the aspects it was lagging behind the most. Whilst AMD has to further push single-threaded performance in the future and continue working on improving memory performance, they’re on Intel’s tail.

The big argument for the 3700X and 3900X is their value as well as their power efficiency. At $329 the 3700X particularly seems exciting and posts near the same gaming performance as the 3900X at $499. Considering that AMD is also shipping the CPU with a viable Wrath Spire, this also adds on to the value that you get if you’re budget conscious.

The 3900X essentially has no competition when it comes to the multi-threaded performance that it’s able to deliver. Here the chip not only bests Intel’s designs, which is able to go toe-to-toe only with >$1500  HEDT platforms, but also suddenly makes AMD’s own Threadripper line-up quite irrelevant.

All in all, while AMD still has some ways to go, they’ve never been this close to Intel in over a decade, and if the company is able to continue to execute as well, we should be seeing exciting things in the future.

Let's block ads! (Why?)

https://www.anandtech.com/show/14605/the-ryzen-3700x-3900x-review-raising-the-bar

2019-07-07 13:05:22Z
52780327326446

AMD Radeon RX 5700 XT and Radeon RX 5700 Review: New Prices Keep Navi In The Game - Tom's Hardware

[unable to retrieve full-text content]

1. AMD Radeon RX 5700 XT and Radeon RX 5700 Review: New Prices Keep Navi In The Game  Tom's Hardware
2. AMD Radeon 5700 Series and the Perils of Pre-Launch Price Cuts  Wccftech
3. NVIDIA's "Super" GPUs Won't Boost Its Gaming Revenue  Motley Fool
4. The Morning After: AMD's pre-release Radeon RX 5700 price drop  Engadget
5. AMD Radeon RX 5700/ RX 5700 XT review: head-to-head with Nvidia Super  Eurogamer.net
6. View full coverage on Google News

https://www.tomshardware.com/reviews/amd-radeon-rx_5700-rx_5700_xt,6216.html

2019-07-07 13:01:50Z
52780327749348

Guidemaster: The best fitness trackers you can buy in 2019 - Ars Technica

Enlarge / A different band can change the entire look of the Inspire HR.

Valentina Palladino

Update: Our Fitness Tracker Guidemaster was originally published in January 2018, but we've been testing new devices to prepare for the beaches and pools of summer 2019. Our recommendations now include some of the best and newest devices available in 2019.

The smartwatch hasn't swallowed up the fitness tracker yet. While many consumers are intrigued by the Apple Watch, Android Wear devices, and the like, old-school fitness trackers can still be useful and available for the right price. The main goal of these devices remains simply tracking activity: from daily movement to intense exercise to steps, heart rate, and sleep. Most of today's fitness trackers haven't changed much aesthetically, either. They're still, by and large, wristbands.

Most modern fitness trackers are meant to be worn all day long. And many now have basic "smartwatch" features, so you don't have to fully sacrifice if you're primarily looking for a wearable to help you get in shape.

With so many devices sharing the same basic goals and set of features, it can be hard to decipher which tracker is right for you. But from our testing, there are some fitness trackers that stand out among the rest—some for their thoughtful applications, others for their versatility, and some for their focused approach to fitness training. So with spring on the horizon and 2018 resolutions still holding strong, we've looked back at the fitness trackers we've reviewed recently and selected the best ones for all kinds of users.

Note: Ars Technica may earn compensation for sales from links on this post through affiliate programs.

Table of Contents

• Best overall
• Fitbit Inspire HR
• The Good
• The Bad
• Runner up
• Fitbit Charge 3
• The Good
• The Bad
• Best for gym-goers
• Garmin Vivosmart 4
• The Good
• The Bad
• Best for newbies
• Moov Now
• The Good
• The Bad
• Most stylish
• Motiv Ring
• The Good
• The Bad

Best overall

Fitbit Inspire HR

Enlarge

Valentina Palladino

Specs at a glance: Fitbit Inspire HR
Price	$99.95
Heart rate monitoring	Yes, continuous
GPS	Connected only
Water resistance	Swimproof
Smartphone alerts	Yes (call, text, and calendar)
Sizes	One size (includes small and large bands)
Battery life	Five days

While we still have love for the Fitbit Alta HR, our previous favorite, the new Fitbit Inspire HR has replaced it in Fitbit’s lineup. Thankfully, it’s just as good as the Alta HR and comes in at just $99. Almost everything we loved about the Alta HR still stands in the Inspire HR—the tracker with interchangeable bands tracks all-day activity, sleep, continuous heart rate, and workouts including swimming.

Fitbit improved the workout-tracking experience in the Inspire HR by giving it a slightly larger touchscreen than the tap-only screen on the Alta HR. Combined with Fitbit’s refined fitness tracker OS, the Inspire HR feels a bit more smartwatch-like than the Alta HR ever did.

Not only can you pick and choose which exercise you want to track using the touchscreen, but you can also set timers and alarms and choose from a few different watch faces to personalize the device. The device can receive smartphone alerts as well. While the Inspire HR can’t do everything the Fitbit Versa or Versa Lite can do, Fitbit distilled some of the most important smartwatch features down so they could work properly and conveniently on the Inspire HR.

The Inspire HR also has Fitbit’s SmartTrack feature, which automatically tracks certain workouts after a period of time, and its connected GPS feature, which lets you use the band in tandem with your phone’s GPS to map outdoor runs and bike rides. The continuous optical heart rate monitor on the module’s underside not only measures your pulse during workouts, but it also keeps track of it at night and that data helps Fitbit’s software measure your time in various stages of sleep.

While the Alta HR lasted about seven days on a single charge, the Inspire HR lasts around five days. We wish the battery lives were comparable, but five days (with nighttime sleep tracking) is still stellar in comparison to most smartwatches. Fitbit’s software is also top-notch—not only are the Android and iOS mobile apps friendly and easy to use, but the company has added numerous new features over the past year or so including guided workouts with Fitbit Coach, menstrual health tracking, social exercise challenges, and more. We’re still waiting for Apple Health integration, but in the meantime, the $99 Inspire HR remains the best fitness tracker for most people.

The Good

• Solid all-purpose fitness and health tracker at a great price.

The Bad

• No altimeter for tracking floors climbed.

Fitbit Inspire HR

Price: $99.95 at Walmart

Buy

Runner up

Fitbit Charge 3

Enlarge

Valentina Palladino

Specs at a glance: Fitbit Charge 3
Price	$149-$169
Heart rate monitoring	Yes, continuous
GPS	Connected only
Water resistance	Up to 50 meters
Smartphone alerts	Yes
Sizes	One size (includes small and large bands)
Battery life	Seven days

The Fitbit Charge 3 has all of the features that the Inspire HR has, plus a few extras. It's slightly wider than the Inspire HR, but that doesn't make it hard to wear. It tracks all-day activity and sleep comfortably, and uses your input as well as Smart Track technology to record workouts.

In terms of activity, the Charge 3's included altimeter is an important differentiator between it and the Inspire HR. An altimeter allows the Charge 3 to track floors climbed, so if you feel particularly accomplished when you take the stairs instead of the elevator and want your wearable to reflect that effort, the Charge 3 is the better device of the two.

Fitbit also included an SpO2 monitor in the Charge 3 which should track blood oxygen levels and allow Fitbit's software to learn more about your sleeping habits (when Fitbit actually enables the sensor).

If you're willing to pay a bit extra, you can get the Charge 3 Special Edition which includes NFC technology for Fitbit Pay. The company's contactless payment system lets you hold your Charge 3 up to an NFC reader to pay for things like coffee, groceries, and the like. If you're ever out on a run and forgot your wallet, you can still pay for things using Fitbit Pay.

Like the Inspire HR, the Charge 3 also has Fitbit's connected GPS feature so you can map outdoor workouts if you have your phone with you. The gap between the Inspire HR and the Charge 3 isn't a big one, but those that value battery life and want option to get NFC payment tech in their fitness tracker should opt for the Charge 3.

The Good

• Good fitness tracker that tracks floors climbed and has a longer battery life than the Inspire HR.

The Bad

• No on-device music controls.

Fitbit Charge 3

Price: $149.95 at Walmart

Buy

Best for gym-goers

Garmin Vivosmart 4

Enlarge

Valentina Palladino

Specs at a glance: Garmin Vivosmart 4
Price	$129.99
Heart rate monitoring	Yes, continuous
GPS	No
Water resistance	Swim- and shower-resistant
Smartphone alerts	Yes
Sizes	Small/medium, large
Battery life	Five to seven days

Garmin packs a lot of features into its small and fairly affordable devices. The $129 Vivosmart 4 is one of the best examples of this, as it includes nearly all of the essential fitness tracker features in addition to rep counting and SpO2 monitoring.

Rep counting was first introduced in the Vivosmart 3, and it plus Garmin’s exercise recognition feature continue to improve with time. It’s fairly accurate and a convenient feature for those who lift weights frequently or do anything other than running or cycling. Exercise recognition allows the device to identify which exercises you're completing; it can, for example, differentiate dumbbell curls from ab jabs and other moves. It’s still not foolproof, but it’s a great feature to have—and even if it mischaracterizes a particular exercise, you can edit it to the correct move in the Garmin Connect mobile app.

SpO2 monitoring recently made its way to Garmin’s elite wearables in addition to the more affordable Vivosmart 4—and unlike in Fitbit devices, the Vivosmart 4 actually puts this sensor to good use. It measures blood oxygen saturation when you’re asleep, potentially catching breathing irregularities that could be signs of disorders like sleep apnea. If you’re fairly healthy, you won’t see very interesting numbers come out of this sensor. But all it takes is one abnormal measurement for you to be more informed about your overall health.

On top of all that, the Vivosmart 4 can do most of what any entry- to mid-tier Garmin wearable can—which is a lot. Garmin redesigned it so now it’s thinner and lighter than before, and has a more attractive look to it. Its OLED display shows the time, a bunch of daily fitness stats, music controls, and more. It also lasts at least five days on a single charge with SpO2 monitoring turned on, meaning you could get more than a week’s worth of life if you choose to turn that feature off.

The Good

• SpO2 monitoring on an affordable device with a great battery life.

The Bad

• No interchangeable bands.

Garmin Vivosmart 4

Price: $129.99 at Amazon

Buy

Best for newbies

Moov Now

Enlarge / Moov Now

Valentina Palladino

Specs at a glance: Moov Now
Price	$42
Heart rate monitoring	No
GPS	No
Water resistance	Waterproof (can track swimming)
Smartphone alerts	No
Sizes	One size
Battery life	Up to six months (replaceable coin cell battery)

Moov Now is a funny little tracker—not just because it's a quarter-sized motion detector, but because it focuses on something bigger than simply tracking movement. The Moov Now sensor can be worn a number of ways, including on your wrist or on your ankle, and it tracks myriad activities. Indoor and outdoor running, cycling, swimming (yes, it's waterproof), and boxing (yes, one on either wrist can be used together) are only a few of the exercises the Moov Now monitors.

But instead of just recording your movements, the Moov Now uses its mobile app to train you to be better at those exercises. With its voice coach, a small voice comes through your smartphone or connected headphones to tell you which types of punches to throw while boxing or when you're landing too hard while running.

You can set goals or areas of improvement within the app so the voice coach knows what to focus on when you're exercising. After a completed workout, the app will provide you more tips in addition to showing you all your movement data. Moov also makes an HR version of its sensor, meant to be worn in a tiny pocket of a headband resting on your temple, so you can pair both to track movement and heart rate at the same time.

The value of Moov Now is remarkable: $59 gets you a motion sensor with different harnesses so you can wear it a variety of ways, a well-designed companion app, and a voice coach to help you get started with any number of exercises. For someone new to fitness with no idea where to start working out, it doesn't get much better than that. Moov Now gives users the chance to try out different forms of exercise, and it removes any excuse to use a fitness tracker for counting steps alone.

The Good

• App monitors sensor's movements in real time to let voice coach help you get better at different exercises.

The Bad

• Cannot be used to track daily movement.

Moov Now (Black)

Price: $42 at Amazon

Buy

Most stylish

Motiv Ring

Enlarge / Motiv Ring

Valentina Palladino

Specs at a glance: Motiv Ring
Price	$174.99
Heart rate monitoring	Yes
GPS	No
Water resistance	Waterproof up to 5ATM
Smartphone alerts	No
Sizes	Multiple (ring numbered sizes)
Battery life	Three days

If you can't bear to give up more wrist real estate, the Motiv Ring might be a good tracker for you. The $199 titanium ring tracks activity as well as heart rate and is meant to be worn all day and all night. On the ring's underside is a heart rate monitor, and on the surface of the ring are small LED lights that alert you to battery life and syncing status. Otherwise, it's an unassuming, minimalistic ring that appeals to both men and women alike.

While the Motiv Ring tracks daily activity, it focuses on "active minutes" or time during the day when your heart rate is elevated. Motiv counts active minutes as those when your heart rate is at 40 percent or more of your aerobic capacity, so everyone's threshold for active minutes will be different. Motiv's mobile app calculates that so the app and the device can work together to track heart rate and give you credit for those minutes in which your heart is working hard.

The Motiv Ring's unique yet familiar form factor could be enough to persuade some users to give it a shot. The device measures heart rate accurately and is very easy to wear all day long, even while washing hands and during sleep. Thanks to its design, waterproof rating, and three-day battery life, you don't have to think about removing the ring for a few days at a time.

But its focus on active minutes is another factor that might appeal to some users who don't want just another step tracker. By encouraging high-intensity activity, the Motiv Ring can help users achieve better heart health over time.

The Good

• Easy-to-wear ring design with accurate heart rate monitor.

The Bad

• Doesn't automatically track workouts.

Motiv Ring

Price: $174.99 at Amazon

Buy

Let's block ads! (Why?)

https://arstechnica.com/gadgets/2019/07/guidemaster-best-fitness-trackers/

2019-07-07 11:22:00Z
CBMiSmh0dHBzOi8vYXJzdGVjaG5pY2EuY29tL2dhZGdldHMvMjAxOS8wNy9ndWlkZW1hc3Rlci1iZXN0LWZpdG5lc3MtdHJhY2tlcnMv0gEA

Warframe Empyrean gameplay demo: "It's not a f$%&ing expansion, it's a connection" - GamesRadar

The centerpiece of Warframe's fourth annual day-long celebration, TennoCon 2019, was a new 42-minute preview of Empyrean. After debuting the new Warframe intro cinematic live on stage, developer Digital Extremes returned to the combined-arms space combat that it first revealed as Codename: Railjack at TennoCon 2018 and then reintroduced as the new Empyrean expansion at E3 2019. It isn't just about flying big ships in co-op teams - it's the most ambitious new direction that Warframe has taken yet, even if Digital Extremes still isn't ready to put a release date on it. Here are five reasons why.

You can kit out and pilot your own gunship

Warframe lets you build out and customize everything: weapons, pets, hoverboards, even your Frames themselves. Once Empyrean goes live, you'll be able to add your own giant gunship to that list. After building a dry dock in your dojo, you can then create your own base model Railjack ship and customize all of its systems - that includes turning your engines fluorescent pink if you have the notion (and that color unlocked). Then you can take your Railjack into missions across the solar system, with the vessel serving both as a mobile HQ for the mission and a legit ship that you can pilot around asteroids and use to shoot down bogeys.

...or control the systems, FTL-style

Maybe you don't feel like taking control of your ship, or maybe you're a guest on somebody else's and they're hogging the helm. You'll still have plenty to do aboard the S.S. Ninja Togetherness. Jump into a turret and blast asteroids for their precious resources or shoot down Grineer interceptors. Pull up the ship menu to reroute power between systems (and so the ship AI stops screaming at you about pending shutdowns). Head to the resource compactors to make sure your materials farming is as efficient as possible. Or do your best Air Force One impression and tell hostile boarding parties to get off your Railjack with classic, third-person combat.

(Image credit: Digital Extremes)

...or jump into space and hijack enemies

If you can tear yourself away from all the joystick jockeying and turret twirling and system selecting, you could even make an asset of yourself on foot and wing. Players can climb into energy catapults and fling themselves at enemy ships with the aid of their Archwing jetpacks; fly over to a vulnerable section of the enemy vessel and you can commence your own boarding action, which plays out like a standard combat mission. Make your way to the bridge - with the help of some tactical strikes from the Railjack if your friends are feeling helpful - take out the pilot, and you can seize the ship. Drive that sucker like a rental, shoot down more enemies, and don't forget to scuttle your hijacked ride with a few shots to the core before you leave.

It lets you recruit NPCs

Warframe has steadily built in more opportunities to interact with NPCs beyond shooting up cannon fodder and getting orders from talking heads, but it's still mostly transactional in nature: you do this or pay this, I give you that. Empyrean will let you make more in-depth relationships by recruiting NPCs to serve on the crew of your Railjack. The example Digital Extremes showed on stage was a cyborg who could give you a lead on some lost technology out in the Heliosphere. Other NPC crewmates might offer different missions or bonuses, and you're free to recruit (and fire) them from as your needs dictate. Think of it like staffing up your ship in Assassin's Creed: Odyssey, but you don't have to defeat your deckhands in one-on-one combat before they'll join you.

Empyrean finally ties the game together

Years of feature updates have turned Warframe into a dizzyingly broad game, for better and for worse. As just one example, Archwing missions are empowering chunks of high-speed, free-form combat, but they're kind of out there in space compared to the rest of the game (literally). Empyrean does more to unite Warframe's many pieces than anything else in the game's history: the on-foot action, the vehicular combat, and even different missions in different parts of the game. Empyrean is the first place Digital Extremes demonstrated its new Squad Link system, which allows teams in entirely different parts of the game to assist each other by completing parallel objectives. Like, say, a shield generator on Earth that's protecting an enemy flagship but is utterly vulnerable to ground-based assault. Yes, just like the Battle of Endor in Star Wars. As creative director Steve Sinclair succinctly explained, Empyrean is "not a f$&%ing expansion, it's a connection."

Save cash for your Warframe platinum packs by doing some shopping from our list of Amazon Prime Day game deals.

Let's block ads! (Why?)

https://www.gamesradar.com/warframe-empyrean-gameplay/

2019-07-07 06:09:00Z
52780327699373

The surprising story behind the Apple Watch's ECG ability - Engadget

Sponsored Links

NOAH BERGER via Getty Images

Welcome to Hitting the Books. With less than one in five Americans reading just for fun these days, we've done the hard work for you by scouring the internet for the most interesting, thought provoking books on science and technology we can find and delivering an easily digestible nugget of their stories.

Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again
by Eric Topol

---

The Apple Watch produced a seismic shift in the public's acceptance of biometric monitoring. Sure, we've had step counters, heart rate and sleep monitors for years, but the Apple Watch made it hip and cool to do so. In Deep Medicine, author Eric Topol examines how recent advances in AI and machine learning techniques can be leveraged to bring (at least the American) healthcare system out of its current dark age and create a more efficient, more effective system that better serves both its doctors and its patients. In the excerpt below, Topol examines the efforts by startup AliveCor and the Mayo Clinic to cram an ECG's functionality into a wristwatch-sized device without -- and this is the important part -- generating potentially lethal false positive results.

In February 2016, a small start-up company called AliveCor hired Frank Petterson and Simon Prakash, two Googlers with AI expertise, to transform their business of smartphone electrocardiograms (ECG). The company was struggling. They had developed the first smartphone app capable of single-lead ECG, and, by 2015, they were even able to display the ECG on an Apple Watch. The app had a "wow" factor but otherwise seemed to be of little practical value. The company faced an existential threat, despite extensive venture capital investment from Khosla Ventures and others.

But Petterson, Prakash, and their team of only three other AI talents had an ambitious, twofold mission. One objective was to develop an algorithm that would passively detect a heart-rhythm disorder, the other to determine the level of potassium in the blood, simply from the ECG captured by the watch. It wasn't a crazy idea, given whom AliveCor had just hired. Petterson, AliveCor's VP of engineering, is tall, blue-eyed, dark-haired with frontal balding, and, like most engineers, a bit introverted. At Google, he headed up YouTube Live, Gaming, and led engineering for Hangouts. He previously had won an Academy Award and nine feature film credits for his design and development software for movies including the

Transformers, Star Trek, the Harry Potter series, and Avatar. Prakash, the VP of products and design, is not as tall as Petterson, without an Academy Award, but is especially handsome, dark-haired, and brown-eyed, looking like he's right out of a Hollywood movie set. His youthful appearance doesn't jibe with a track record of twenty years of experience in product development, which included leading the Google Glass design project. He also worked at Apple for nine years, directly involved in the development of the first iPhone and iPad. That background might, in retrospect, be considered ironic.

Meanwhile, a team of more than twenty engineers and computer scientists at Apple, located just six miles away, had its sights set on diagnosing atrial fibrillation via their watch. They benefited from Apple's seemingly unlimited resources and strong corporate support: the company's chief operating officer, Jeff Williams, responsible for the Apple Watch development and release, had articulated a strong vision for it as an essential medical device of the future. There wasn't any question about the importance and priority of this project when I had the chance to visit Apple as an advisor and review its progress. It seemed their goal would be a shoo-in.

The Apple goal certainly seemed more attainable on the face of it. Determining the level of potassium in the blood might not be something you would expect to be possible with a watch. But the era of deep learning, as we'll review, has upended a lot of expectations.

The idea to do this didn't come from AliveCor. At the Mayo Clinic, Paul Friedman and his colleagues were busy studying details of a part of an ECG known as the T wave and how it correlated with blood levels of potassium. In medicine, we've known for decades that tall T waves could signify high potassium levels and that a potassium level over 5.0 mEq/L is dangerous. People with kidney disease are at risk for developing these levels of potassium. The higher the blood level over 5, the greater the risk of sudden death due to heart arrhythmias, especially for patients with advanced kidney disease or those who undergo hemodialysis. Friedman's findings were based on correlating the ECG and potassium levels in just twelve patients before, during, and after dialysis. They published their findings in an obscure heart electrophysiology journal in 2015; the paper's subtitle was "Proof of Concept for a Novel 'Blood-Less' Blood Test." They reported that with potassium level changes even in the normal range (3.5–5.0), differences as low as 0.2 mEq/L could be machine detected by the ECG, but not by a human-eye review of the tracing.

Friedman and his team were keen to pursue this idea with the new way of obtaining ECGs, via smartphones or smartwatches, and incorporate AI tools. Instead of approaching big companies such as Medtronic or Apple, they chose to approach AliveCor's CEO, Vic Gundotra, in February 2016, just before Petterson and Prakash had joined. Gundotra is another former Google engineer who told me that he had joined AliveCor because he believed there were many signals waiting to be found in an ECG. Eventually, by year's end, the Mayo Clinic and AliveCor ratified an agreement to move forward together.

The Mayo Clinic has a remarkable number of patients, which gave AliveCor a training set of more than 1.3 million twelve-lead ECGs gathered from more than twenty years of patients, along with corresponding blood potassium levels obtained within one to three hours of the ECG, for developing an algorithm. But when these data were analyzed it was a bust.

Here, the "ground truths," the actual potassium (K+) blood levels, are plotted on the x-axis, while the algorithm-predicted values are on the y-axis. They're all over the place. A true K+ value of nearly 7 was predicted to be 4.5; the error rate was unacceptable. The AliveCor team, having made multiple trips to Rochester, Minnesota, to work with the big dataset, many in the dead of winter, sank into what Gundotra called "three months in the valley of despair" as they tried to figure out what had gone wrong.

Petterson and Prakash and their team dissected the data. At first, they thought it was likely a postmortem autopsy, until they had an idea for a potential comeback. The Mayo Clinic had filtered its massive ECG database to provide only outpatients, which skewed the sample to healthier individuals and, as you would expect for people walking around, a fairly limited number with high potassium levels. What if all the patients who were hospitalized at the time were analyzed? Not only would this yield a higher proportion of people with high potassium levels, but the blood levels would have been taken closer to the time of the ECG.

They also thought that maybe all the key information was not in the T wave, as Friedman's team had thought. So why not analyze the whole ECG signal and override the human assumption that all the useful information would have been encoded in the T wave? They asked the Mayo Clinic to come up with a better, broader dataset to work with. And Mayo came through. Now their algorithm could be tested with 2.8 million ECGs incorporating the whole ECG pattern instead of just the T wave with 4.28 million potassium levels. And what happened?

The receiver operating characteristic (ROC) curves of true versus false positive rates, with examples of worthless, good, and excellent plotted. Source: Wikipedia (2018)

Eureka! The error rate dropped to 1 percent, and the receiver operating characteristic (ROC) curve, a measure of predictive accuracy where 1.0 is perfect, rose from 0.63 at the time of the scatterplot to 0.86. We'll be referring to ROC curves a lot throughout the book, since they are considered one of the best ways to show (underscoring one, and to point out the method has been sharply criticized and there are ongoing efforts to develop better performance metrics) and quantify accuracy—plotting the true positive rate against the false positive rate (Figure 4.2). The value denoting accuracy is the area under the curve, whereby 1.0 is perfect, 0.50 is the diagonal line "worthless," the equivalent of a coin toss. The area of 0.63 that AliveCor initially obtained is deemed poor. Generally, 0.80–.90 is considered good, 0.70–.80 fair. They further prospectively validated their algorithm in forty dialysis patients with simultaneous ECGs and potassium levels. AliveCor now had the data and algorithm to present to the FDA to get clearance to market the algorithm for detecting high potassium levels on a smartwatch.

There were vital lessons in AliveCor's experience for anyone seeking to apply AI to medicine. When I asked Petterson what he learned, he said, "Don't filter the data too early. . . . I was at Google. Vic was at Google. Simon was at Google. We have learned this lesson before, but sometimes you have to learn the lesson multiple times. Machine learning tends to work best if you give it enough data and the rawest data you can. Because if you have enough of it, then it should be able to filter out the noise by itself."

"In medicine, you tend not to have enough. This is not search queries. There's not a billion of them coming in every minute. . . . When you have a dataset of a million entries in medicine, it's a giant dataset. And so, the order or magnitude that Google works at is not just a thousand times bigger but a million times bigger." Filtering the data so that a person can manually annotate it is a terrible idea. Most AI applications in medicine don't recognize that, but, he told me, "That's kind of a seismic shift that I think needs to come to this industry."

Excerpted from Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again. Copyright © 2019 by Eric Topol. Available from Basic Books.

In this article: apple, column, gadgetry, gadgets, google, tomorrow

All products recommended by Engadget are selected by our editorial team, independent of our parent company. Some of our stories include affiliate links. If you buy something through one of these links, we may earn an affiliate commission.

137 Shares

Share

Tweet

Share

Save

Let's block ads! (Why?)

https://www.engadget.com/2019/07/06/the-surprising-story-behind-apple-watchs-ecg-function/

2019-07-06 16:56:55Z
CAIiEGI6sMyPPtkO_aZnxwW-4YcqFwgEKg8IACoHCAowwOjjAjDp3xsw9bAl

The surprising story behind the Apple Watch's ECG ability - Engadget

Sponsored Links

NOAH BERGER via Getty Images

Welcome to Hitting the Books. With less than one in five Americans reading just for fun these days, we've done the hard work for you by scouring the internet for the most interesting, thought provoking books on science and technology we can find and delivering an easily digestible nugget of their stories.

Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again
by Eric Topol

---

The Apple Watch produced a seismic shift in the public's acceptance of biometric monitoring. Sure, we've had step counters, heart rate and sleep monitors for years, but the Apple Watch made it hip and cool to do so. In Deep Medicine, author Eric Topol examines how recent advances in AI and machine learning techniques can be leveraged to bring (at least the American) healthcare system out of its current dark age and create a more efficient, more effective system that better serves both its doctors and its patients. In the excerpt below, Topol examines the efforts by startup AliveCor and the Mayo Clinic to cram an ECG's functionality into a wristwatch-sized device without -- and this is the important part -- generating potentially lethal false positive results.

In February 2016, a small start-up company called AliveCor hired Frank Petterson and Simon Prakash, two Googlers with AI expertise, to transform their business of smartphone electrocardiograms (ECG). The company was struggling. They had developed the first smartphone app capable of single-lead ECG, and, by 2015, they were even able to display the ECG on an Apple Watch. The app had a "wow" factor but otherwise seemed to be of little practical value. The company faced an existential threat, despite extensive venture capital investment from Khosla Ventures and others.

But Petterson, Prakash, and their team of only three other AI talents had an ambitious, twofold mission. One objective was to develop an algorithm that would passively detect a heart-rhythm disorder, the other to determine the level of potassium in the blood, simply from the ECG captured by the watch. It wasn't a crazy idea, given whom AliveCor had just hired. Petterson, AliveCor's VP of engineering, is tall, blue-eyed, dark-haired with frontal balding, and, like most engineers, a bit introverted. At Google, he headed up YouTube Live, Gaming, and led engineering for Hangouts. He previously had won an Academy Award and nine feature film credits for his design and development software for movies including the

Transformers, Star Trek, the Harry Potter series, and Avatar. Prakash, the VP of products and design, is not as tall as Petterson, without an Academy Award, but is especially handsome, dark-haired, and brown-eyed, looking like he's right out of a Hollywood movie set. His youthful appearance doesn't jibe with a track record of twenty years of experience in product development, which included leading the Google Glass design project. He also worked at Apple for nine years, directly involved in the development of the first iPhone and iPad. That background might, in retrospect, be considered ironic.

Meanwhile, a team of more than twenty engineers and computer scientists at Apple, located just six miles away, had its sights set on diagnosing atrial fibrillation via their watch. They benefited from Apple's seemingly unlimited resources and strong corporate support: the company's chief operating officer, Jeff Williams, responsible for the Apple Watch development and release, had articulated a strong vision for it as an essential medical device of the future. There wasn't any question about the importance and priority of this project when I had the chance to visit Apple as an advisor and review its progress. It seemed their goal would be a shoo-in.

The Apple goal certainly seemed more attainable on the face of it. Determining the level of potassium in the blood might not be something you would expect to be possible with a watch. But the era of deep learning, as we'll review, has upended a lot of expectations.

The idea to do this didn't come from AliveCor. At the Mayo Clinic, Paul Friedman and his colleagues were busy studying details of a part of an ECG known as the T wave and how it correlated with blood levels of potassium. In medicine, we've known for decades that tall T waves could signify high potassium levels and that a potassium level over 5.0 mEq/L is dangerous. People with kidney disease are at risk for developing these levels of potassium. The higher the blood level over 5, the greater the risk of sudden death due to heart arrhythmias, especially for patients with advanced kidney disease or those who undergo hemodialysis. Friedman's findings were based on correlating the ECG and potassium levels in just twelve patients before, during, and after dialysis. They published their findings in an obscure heart electrophysiology journal in 2015; the paper's subtitle was "Proof of Concept for a Novel 'Blood-Less' Blood Test." They reported that with potassium level changes even in the normal range (3.5–5.0), differences as low as 0.2 mEq/L could be machine detected by the ECG, but not by a human-eye review of the tracing.

Friedman and his team were keen to pursue this idea with the new way of obtaining ECGs, via smartphones or smartwatches, and incorporate AI tools. Instead of approaching big companies such as Medtronic or Apple, they chose to approach AliveCor's CEO, Vic Gundotra, in February 2016, just before Petterson and Prakash had joined. Gundotra is another former Google engineer who told me that he had joined AliveCor because he believed there were many signals waiting to be found in an ECG. Eventually, by year's end, the Mayo Clinic and AliveCor ratified an agreement to move forward together.

The Mayo Clinic has a remarkable number of patients, which gave AliveCor a training set of more than 1.3 million twelve-lead ECGs gathered from more than twenty years of patients, along with corresponding blood potassium levels obtained within one to three hours of the ECG, for developing an algorithm. But when these data were analyzed it was a bust.

Here, the "ground truths," the actual potassium (K+) blood levels, are plotted on the x-axis, while the algorithm-predicted values are on the y-axis. They're all over the place. A true K+ value of nearly 7 was predicted to be 4.5; the error rate was unacceptable. The AliveCor team, having made multiple trips to Rochester, Minnesota, to work with the big dataset, many in the dead of winter, sank into what Gundotra called "three months in the valley of despair" as they tried to figure out what had gone wrong.

Petterson and Prakash and their team dissected the data. At first, they thought it was likely a postmortem autopsy, until they had an idea for a potential comeback. The Mayo Clinic had filtered its massive ECG database to provide only outpatients, which skewed the sample to healthier individuals and, as you would expect for people walking around, a fairly limited number with high potassium levels. What if all the patients who were hospitalized at the time were analyzed? Not only would this yield a higher proportion of people with high potassium levels, but the blood levels would have been taken closer to the time of the ECG.

They also thought that maybe all the key information was not in the T wave, as Friedman's team had thought. So why not analyze the whole ECG signal and override the human assumption that all the useful information would have been encoded in the T wave? They asked the Mayo Clinic to come up with a better, broader dataset to work with. And Mayo came through. Now their algorithm could be tested with 2.8 million ECGs incorporating the whole ECG pattern instead of just the T wave with 4.28 million potassium levels. And what happened?

The receiver operating characteristic (ROC) curves of true versus false positive rates, with examples of worthless, good, and excellent plotted. Source: Wikipedia (2018)

Eureka! The error rate dropped to 1 percent, and the receiver operating characteristic (ROC) curve, a measure of predictive accuracy where 1.0 is perfect, rose from 0.63 at the time of the scatterplot to 0.86. We'll be referring to ROC curves a lot throughout the book, since they are considered one of the best ways to show (underscoring one, and to point out the method has been sharply criticized and there are ongoing efforts to develop better performance metrics) and quantify accuracy—plotting the true positive rate against the false positive rate (Figure 4.2). The value denoting accuracy is the area under the curve, whereby 1.0 is perfect, 0.50 is the diagonal line "worthless," the equivalent of a coin toss. The area of 0.63 that AliveCor initially obtained is deemed poor. Generally, 0.80–.90 is considered good, 0.70–.80 fair. They further prospectively validated their algorithm in forty dialysis patients with simultaneous ECGs and potassium levels. AliveCor now had the data and algorithm to present to the FDA to get clearance to market the algorithm for detecting high potassium levels on a smartwatch.

There were vital lessons in AliveCor's experience for anyone seeking to apply AI to medicine. When I asked Petterson what he learned, he said, "Don't filter the data too early. . . . I was at Google. Vic was at Google. Simon was at Google. We have learned this lesson before, but sometimes you have to learn the lesson multiple times. Machine learning tends to work best if you give it enough data and the rawest data you can. Because if you have enough of it, then it should be able to filter out the noise by itself."

"In medicine, you tend not to have enough. This is not search queries. There's not a billion of them coming in every minute. . . . When you have a dataset of a million entries in medicine, it's a giant dataset. And so, the order or magnitude that Google works at is not just a thousand times bigger but a million times bigger." Filtering the data so that a person can manually annotate it is a terrible idea. Most AI applications in medicine don't recognize that, but, he told me, "That's kind of a seismic shift that I think needs to come to this industry."

Excerpted from Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again. Copyright © 2019 by Eric Topol. Available from Basic Books.

In this article: apple, column, gadgetry, gadgets, google, tomorrow

All products recommended by Engadget are selected by our editorial team, independent of our parent company. Some of our stories include affiliate links. If you buy something through one of these links, we may earn an affiliate commission.

137 Shares

Share

Tweet

Share

Save

Let's block ads! (Why?)

https://www.engadget.com/2019/07/06/the-surprising-story-behind-apple-watchs-ecg-function/

2019-07-06 16:53:10Z
CAIiEGI6sMyPPtkO_aZnxwW-4YcqFwgEKg8IACoHCAowwOjjAjDp3xsw9bAl