Ryzen 9 3950X: AMD's 16-core CPUs is an epic end-zone dance over Intel
- 15 November, 2019 01:00
Conclusion: The 16-core Ryzen 9 3950X is simply the fastest consumer CPU in town. Period. Full stop. Review over.
While that may seem like the most unorthodox way to start a review of AMD’s new 16-core Ryzen 9 3950X, anybody expecting a different outcome at this point was just fooling themselves.
Most rational people can admit that the game was essentially over once AMD’s Zen 2-based Ryzen 3000 chips hit the shelves in the 6- to 12-core range. Sure, Intel had a claim to gaming performance, but for many people it probably wasn’t enough of a justification.
With the 16-core Ryzen 9 3950X, AMD is essentially running down the field, spiking the ball, and doing what the NFL would probably fine for excessive end-zone celebrations these days. Yes, fans wearing blue jerseys may consider heading for the parking lot or simply waiting around so you can end up on the Jumbotron tearfully watching the clock run out.
This is the moment for red jersey fans, though, and you’ll still be cheering hoarsely as the Commissioner clears the field for the awards ceremony. So stick around as we break down just how insanely, stupidly fast the new Ryzen 9 3950X is.
What is Ryzen 9 3950X?
If you read the news of the Zen 2 from last year or our review of the 12-core Ryzen 9 3900X, you already know what you need to know about AMD’s Ryzen 3000 CPUs. With the 16-core Ryzen 9 3950X, AMD essentially takes the cream of the crop of Zen 2 cores (what’s called binning) and drops the very best chips into the Ryzen 9 3950X. Obviously the CPU has four more cores than the 12-core Ryzen 9 3900X. Clocks also vary, with the Ryzen 9 3950X boosting to 4.7GHz vs. the 4.6GHz of the Ryzen 9 3900X. Base clocks—essentially worst-case-scenario clock speeds—are also lower, at 3.5GHz in the 3950X vs 3.8GHz in the 3900X. That’s to be expected, as more cores means more heat to dissipate.
Little else changes except, well, the price, which some diehards consider too high (we disagree). and the performance, which is everything you’ve come to expect of AMD’s Zen 2 cores.
How we tested
For our review, we used the same MSI X570 Godlike motherboard we’d used with the Ryzen 9 3900X, with 16GB of DDR4/3600 CL16 in dual-channel mode, Windows 10 1903 and a Founders Edition GeForce RTX 2080 Ti card. As the AMD platform supports PCIe 4.0, we used a 2TB Corsair MP600 PCIe 4.0.
For the Intel system, we used an Asus Z390 ROG Maximus XI Hero WiFi to test the Core i9-9900KS and the Core i9-9900K chip. For storage, the Intel chips were tested with a 512GB Samsung 960 Pro drive, while the GPU as well as Nvidia drivers were the same. It also had 16GB of DDR4 in dual-channel mode but slightly lower clocks at DDR4/3400 CL16.
While the original Ryzen 9 3900X, Core i9-9900KS and Core i9-9900K were tested with matching Corsair H80 V2’s with manual 100 percent fans, we used the supplied NZXT Kraken X62 280mm cooler for the Ryzen 9 3950X. Although we didn’t retest the previous CPUs with the same cooler for time’s sake, to be fair to Intel we will revisit its chip with the slightly beefier cooling. However, our numbers essentially matched the company’s sanity-check scores, so we don’t expect the results to drastically change. And yes AMD fans used to that freebie cooler, you don’t get it with this CPU despite the higher price.
We decided with the Ryzen 9 3900X to give up on the argument over MCE Auto vs. PBO Auto on for testing. Our rationale is that it comes switched on out of the box by all board vendors. Testing with it off is further from fact than testing with it on auto, even if auto may vary slightly from board maker to board maker.
One other note: For our test we stuck to decidedly more “consumer” platforms. AMD likes to compare the Ryzen 9 3950X to the pricier $1,189 12-core Core i9-9920X platform. We think the better comparison is with the more affordable AM4/LGA1151 socket platforms, where motherboards can be had for far less cash.
Ryzen 9 3950X 3D rendering performance
We’ll kick off our testing with Maxon’s current 3D rendering benchmark, Cinebench R20. It measures CPU performance using the same engine used in Maxon’s Cinema4D. Besides being a real product, Cinema4D is also integrated into effects engines such as that small indie app called Adobe After Effects.
First up is multi-threaded performance, where the result is pretty much everything you expected to see from a 16-core CPU going up against 12-core and 8-core CPUs. The Ryzen 9 3950X simply demolishes Intel’s best “consumer” CPU.
We also use Cinebench to measure single-threaded performance, to give you an idea of how it will perform in far more mundane applications such as the majority of Office tasks and even Adobe Photoshop. Here’s the shocker: The 4.7GHz Ryzen 9 3950X slightly edges the 5GHz Core i9 chip. It’s close enough you wouldn’t be able to tell, but to see the Ryzen 9 3950X even slightly outscore the 5GHz Core i9-9900KS was unexpected.
We also looked at performance of the CPUs using an oldie but goodie, the POV Ray 3.7 benchmark. This ray tracing application has roots in the Amiga era, but it's been dutifully updated by supporters. The multi-core benchmark again isn’t much of a surprise as the Ryzen 9 3950X runs down the field unopposed and spikes the ball—again.
We also ran POV Ray 3.7 using a single thread. Unlike with Cinebench R20, both Ryzen’s slightly trail the Intel CPUs. It’s still fairly close but clearly if you had to pick a winner, it would be Core i9.
Our next 3D rendering test is Chaosgroup’s V-Ray Next 4.2. V-Ray is a physically based renderer that has been used on such little projects as Avengers: Endgame, where V-Ray was used on the Thanos character, among other Marvel-related things. The result is as expected: a thrashing of the Core i9 at the hands of Thanos. And yes, all that for a drop of blood.
Our last rendering test is the Corona Renderer. Corona is an unbiased photorealistic renderer—in other words, the calculations used to render the images doesn’t take short cuts.
One thing Corona is particularly proud of is being CPU-based only. You can read the company’s blog post on why it sticks to CPUs for rendering. You should note that it has embraced Intel’s Embree for ray tracing. And then you can look in awe at the performance of the 16-core Ryzen 9 3950X.
Ryzen 9 3950X Encoding Performance
The results above are all expected. Of the tasks typically done on computer, 3D rendering and 3D modelling usually offer the best performance scaling for the CPU cores.
Video encoding, however, is a little more tricky. Memory bandwidth, special instructions in the CPU, and the microarchitecture, as well as the GPU, can factor into into the end result.
For our first test, we used the free, popular, and very efficient HandBrake app to convert a 4K version of the open-source Tears of Steel video to 1080p using HEVC. The version we use was downloaded on October 20.
HandBrake generally favors more cores (up to a certain point), and the 16-core Ryzen 9 3950X again walks away from the crowd. Interestingly, the 5GHz Core i9-9900KS comes in second, beating the 12-core Ryzen 9 3900X, while the Core i9-9900K comes last.
It’s hard to say why, as the Core i9-9900KS is essentially the same as the Core i9-9900K except with higher clocks speeds most of the time. We suspect that may be the case here, as the 16-minute workload might be short enough for thermals not to be an issue.
To take it up a notch, we also tasked the various CPUs to take the same 4K video and convert it using HEVC, but using an output at the same resolution but a higher frame rate. One thing this does is almost triple the time it takes. We now see the Core i9-9900KS slightly slower than the Core i9-9900K, which we suspect is the result of running flat-out for almost 45 minutes. We’ll have to revisit it down the road with beefier cooling.
One last thing we should note is Intel’s Quick Sync performance, which we didn’t present here. Intel CPUs with integrated graphics have hardware circuits dedicated to video encoding and decoding, and they simply sing when used. For example, the Intel IGP in the Core i9-9900K (which is the same as the Core i9-9900KS) will take half the encode the same video as the 16-core Ryzen 9 3950X.
Our next encoding test is Cinegy’s Cinescore 10.4. It’s a free benchmark that tests various codecs at various resolutions. It's designed to fit in memory and make it easy for broadcast industry experts to measure the performance of off-the-shelf hardware.
The clear winner is again Ryzen 9 3950X, but we will note the margins, though quite good, don’t scale with core count. As we’ve said before, video encoding doesn’t always scale with core count the way you’d expect.
Our last video encoding test uses Adobe’s Premiere CC to export a short 4K video shot on Sony Alpha mirrorless cameras, using the Blu-ray preset with the Maximum Render option checked off. That options is supposed to improve visual quality when resolutions are changed in a project. We also render the project directly to the same Plextor PCIe drive, and instead of using the GeForce RTX 2080 Ti to encode, we use the CPU.
The winner, by a mile, is the 16-core Ryzen 9 3950X, taking about 40 percent less time. Again, you don’t see the scaling as you with 3D modelling, but it’s clear who won this.
Because we know most people who paid for a GPU will actually use it, we also run the same export, but now using the GeForce RTX 2080 Ti CUDA engine for the encode. The result puts most of the work on the GPU, but the CPU still matters. It looks like higher clock speeds take the day here, as the Core i9-9900KS slightly edges the Core i9-9900K and the Ryzen 9 3950X.
One thing we know about Adobe Premiere CC (and a lot of video encoding tasks) is that performance results will vary depending on which codec you use, as well as which settings you pick for an encode. For this last test, we take the same video but export a 4K video using the HEVC High profile within Premiere. The Ryzen 9 3950X comes in first again, but we’re down to about a 16-percent advantage over the 8-core Intel CPU.
Keep reading for more benchmarks, including gaming.
Ryzen 9 3950X Compression Decompression Performance
We’ll close off application performance by looking at the compression and decompression of the chips. The first result uses the built-in benchmark inside of RARLab’s WinRAR. WinRAR has traditionally not favored AMD’s Ryzen, but the winner through sheer will power is the Ryzen 9 3950X.
The thing is, WinRAR is not free, and free sells. So, our next test uses the popular 7-Zip compression/decompression utility. Looking at the built-in benchmark for the compression score, it’s all win for Ryzen over Intel. 7-Zip’s decompression test is mostly dependent on integer performance and branch prediction capability—all things a 16-core CPU has in spades, as you can see by the results. The benchmark is multi-threaded as well, so the results are not unexpected.
One potential issue for Ryzen, though, is 7-Zip’s Compression test. This portion is more memory-dependent. Cache performance, memory latency, and out-of-order performance matters, 7-Zip creator’s say. While both Ryzens easily ace the Intel parts, the 16-core Ryzen 9 is just about dead even with the 12-core Ryzen 9 CPU. Because the test is multi-threaded, we would expect the 16-core also to outperform the 12-core chip easily, but it doesn’t. We wonder whether the dual-channel memory of the Ryzen is to blame. But hey, it’s still a big win.
Ryzen 9 3950X Gaming Performance
If you’re buying a 16-core CPU solely to play games, you might want to consider an 8-core chip, or even a 6-core chip instead, because few games can use all of the cores. That’s the conventional wisdom, anyway. The problem is modern gaming isn’t about playing a game, winning, and going to sleep warm and happy. It’s about streaming it live, or recording it while adding LOL memes, sound effects, and all kinds of “gaming”-related tasks that didn't exist a few years ago.
In theory, with its slightly higher boost speeds and “best of the best” dies, the 16-core Ryzen 9 3950X should be AMD’s ultimate gaming CPU. For the most part, it is.
For our gaming section we tested only at 1920x1080 resolution. You can get 300MHz monitors now (at 1920x1080) and at anything higher-res, even with the mighty GeForce RTX 2080 Ti, it’s pretty much dead-even almost all of the time. And yeah, turn on ray tracing options, and you are 100 percent GPU-bound all of the time at any resolution.
First up is Shadows of the Tomb Raider, which we ran at 1920x1080 resolution using the Highest quality preset, but with ray tracing off. This is likely the only good news (although it’s not a lot) for Intel. The higher clock speed and other Intel magic still rings true in game. Unfortunately, it’s only about 10 percent in Shadows of the Tomb Raider and at 1920x1080, which makes the game less bottlenecked by the CPU.
The Ryzen 9 3950X and its teammate, the Ryzen 9 3900X, close up a little more in Deus Ex: Mankind Divided, where they trail by about 8 percent. A loss, but AMD didn’t plan on winning in gaming, it just wanted to close the gap.
In the newly released Gears of War 5 benchmark, it’s a similar 6- to 7-percent performance gap between the Ryzens and Core i9s. It’s about a 6- to 7-percent clock speed difference between them too.
One inconvenient truth: Some things just run better on Intel chips. We see that illustrated in Far Cry 5 run on Ultra at 1920x1080 resolution: The familiar “Ryzen gaming gap” of 20 to 23 percent rears its head. Even at 2560x1440 resolution, Intel’s chips have the edge, although the gap closes to 13 percent.
There are exceptions to this rule. One is Ashes of the Singularity: Escalation, the original poster child for multi-core support in DX12. For the most part, it’s a dead tie when looking at the CPU heavy preset, which throws more units on the screen. Technically the Ryzen 9 3950X lands in front, but we consider them mostly even.
While new games are typically bottlenecked by the graphics card, the vast majority of games people play are not. These eSports-grade games typically run on fairly low-end hardware, which moves the needle further toward the CPU. One such game is the popular Rainbow Six Siege. We run it at 1920x1080 on Ultra, and predictably the Intel chips win—but the gap is fairly small at 10 percent, and all of the CPUs are kicking out in excess of 300 fps.
We’ll close out gaming with the immensely popular Counter Strike: Global Operations at 1920x1080 resolution set to high quality. We run the FPS Workshop map for our results, which features both interior and exterior scenes. An excessive amount of smoke particle effects at the end greatly drags down frame rates.
The winner is Intel by a healthy 14 percent. But as with Rainbow Six Siege, we're looking at frame rates in excess of 400 fps to 500 fps for all CPUs involved. So outside of a professional player who really does need the excessive frame rates, it’s not much of a victory.
Ryzen 9 3950X Performance analysis
We’ll close this out by trying to give you an idea of the strengths of the new Ryzen 9 3950X across various workloads. For this test, we use Cinebench R15 and task the CPU to cough up a score using from 1 thread to the maximum amount of threads the CPU has.
For our first chart, we compare the scores of the 16-core Ryzen 9 3950X with the 12-core Ryzen 9 3900X. You can see that in the chart below.
It’s a little hard to visualize the differences in the chart above, so we also graph out the performance difference between the chips. As you can see, the Ryzen 9 3950X wins against the already stupidly good Ryzen 9 3900X by giving us single-digit increases in performance on lightly threaded loads, then double-digit jumps up to 30 percent when all threads are loaded up. That’s simply stunning.
Perhaps more important is how the 16-core Ryzen 9 3950X compares to Intel’s shiny new 8-core Core i9-9900KS. Sure Intel fans will say they don’t compare, but both are the alpha CPUs in their sockets.
Looking at it from a percent difference again helps visualize the stunning performance differences between these two CPUs. The left side of the chart shows a little bit of a negative for the Ryzen 9 vs. the Core i9, but we are talking single-digit differences. Maybe that depression is a road bump for the Ryzen 9 3950X, as it hurtles forward with up to an 88-percent advantage over the Intel CPU.
The performance we’re seeing in the Ryzen 9 3950X is truly historic for consumers.
Yes, some will complain that the $749 launch price for the Ryzen 9 3950X is too high, but when you remember the tears of joy nerds shed when the 16-core Threadripper 1950X made its debut in July of 2017 for $999, and then think about the 16-core Threadripper 2950X for $899—and don't forget Intel’s $1,699 Core i9-9960X from 2018—this actually feels like a deal for this insane amount of performance.
Yes, Intel has cut the price of the upcoming 18-core Core i9-10980XE to $999, but after seeing the performance of the Ryzen 9 3950X, we’re not sure it’s enough.