Yesterday we took a closer look at the new Threadripper 2990WX and 2950X processors. In short, the 2950X impressed with solid performance across the board at a premium yet competitive price. The 2990WX was less impressive with mixed performance and a high price.
At the moment, Windows 10 is blamed for the mixed results, and we may need to investigate further and keep our fingers crossed for an update. In the meantime, a large number of people raised concerns about our and many other tests as we only tested one application at a time.
We always planned to create a separate, multitasking benchmark, and we said that in yesterday's review. While we were working on and completing the 2990WX and 2950X tests, we looked at multitasking performance. After further test days, we have a few more results to discuss.
For this test, the 2990WX was compared directly to the Core i9-7980XE and we need to review 5 batches of multitasking benchmark results. All tests were performed using standard settings with a 240mm closed loop cooler on each CPU. The Wraith Ripper jumped up and down, causing the 2990WX to throttle in our benchmarks, even though we showed footage that contradicts that. However, it was not throttled and we tested both air and liquid while creating the original content and found no difference in inventory settings.
Later, both systems were configured with 64 GB DDR4 memory using eight 8 GB modules. This limited the 2990WX to DDR4-3000, while the 7980XE could be much higher, but we chose DDR4-3200 memory with low latency to try to keep things even and somewhat realistic.
The following diagrams may look a bit strange at first glance, especially compared to what you normally see. However, we hope that the format is easy to follow as you get all the information on a single chart per test.
First, we'll look at how the 2990WX and 7980XE compare when running Blender Open Data and 7-zip. Note that the results here show how the CPUs work individually in each application. That said, Blender was running and completed before testing 7-Zip continued. So these are just standard benchmark results.
Here the 2990WX did the BMW27 workload 36% faster and the Barbershop workload 25% faster. Then we see with 7-Zip that the 2990WX is 116% fiber for decompression work, but 35% slower for compression work. Now let's run both the Blender and 7-Zip applications at the same time. 7-Zip is in an endless loop while Blender is running the standard workloads.
Here we see some interesting things. The 2990WX is now 58% faster for the BMW27 workload and 30% faster for the Barbershop workload. However, the 7-Zip performance is less affordable. While the 2990WX was 116% faster in decompression performance, it is now 3% slower. It was 35% slower in the compression test and now only 16% slower.
Basically, the 2990WX was already very impressive in the Blender rendering test, and adding 7-Zip to the mix only makes it look more impressive, but the huge advantage it had in 7-Zip is now gone and does it both in compression as well as in decompression inferior tests. The 7980XE seems to prioritize 7-Zip while the 2990WX blender prioritizes.
Next we test with HandBrake and RealBench. For HandBrake we have an H.264 4K 60fps video that we convert to H.265 4K 60 fps and we give the average frame rate and total render time. Then RealBench will loop its heavy multitasking workload.
The 2990WX doesn't do well in HandBrake and was 31% slower than the 7980XE. It's not particularly good in RealBench either, as it takes 10% longer to complete a single pass. Now let's combine these two tests by running them at the same time.
When both applications are running at the same time, the 2990WX is now 17% slower than the 7980XE in HandBrake, which is a reasonable improvement, but still slower, of course. While we saw a reduced margin in HandBrake, we see the margin increase slightly for RealBench, the 2990WX is now 15% slower. Therefore, the 2990WX could not really keep up in this multitasking benchmark on the 7980XE.
One of the worst results for the 2990XW seen in our first cover was found when testing with VeraCrypt, 1 GB test. We combine this test with one of the best test results for the 32-core processor that was seen in Corona. Here the 2990WX was 28% faster than the 7980XE and as we have seen it is always good for rendering workloads. For VeraCrypt, however, it was 47% slower, not a great result. Now let's run these applications at the same time.
As we saw in the first set of multitasking benchmarks, AMD seems to prioritize rendering workloads, while Intel prefers memory-sensitive workloads. The 7980XE took almost 6x longer than the 2990WX to meet the Corona benchmark, while decryption performance was only reduced by 7%. Meanwhile, the 2990WX reduced decryption performance by 34% and rendering time by 61%.
The 32-core processor appears more balanced here and should be equipped with 14 additional cores. Still, we have mixed results for the 2990WX. It's amazing in Corona and awful in VeraCrypt. This applies regardless of whether you run these applications individually or simultaneously.
This time we have three applications: HandBrake, CPUz and 7-Zip. So they all work individually on each CPU. The 2990WX is 22% slower in HandBrake, 95% faster in CPUz, and again we see the same mixed 7-Zip performance that we discussed earlier.
Running all three applications simultaneously gives some interesting results. While the 2990WX in HandBrake was 22% slower before, it's now 29% faster, which is impressive. CPUz was a huge success, but was still 21% faster than the 7980XE.
Even in 7-Zip we find positive results for the 32-core processor. He managed to maintain a good deal of his decompression performance, and the 7980XE scored a big enough hit in the compression test that the 2990WX had only a head start. This is the first really great result we saw for the 2990WX in these multitasking benchmarks, as we didn't include a rendering workload to support the 32-core processor.
Finally, we have another triple header with POV-Ray, WinRAR and F1 2017. POV-Ray gave the 2990WX the best result in our first test and as you can see it is 36% faster in this test than the 7980XE. While WinRAR is not very core-intensive, it is extremely memory-intensive, so it will be interesting to see if this affects POV-Ray performance.
The 2990WX was 59% slower than the 7980XE when tested with WinRAR, or in other words, the 7980XE was 142% faster. Finally we have F1 2017 and here the 2990WX was 24% slower compared to the average frame rate and 30% slower for the 1% low result.
As we have seen over and over again, multitasking only extends the advantage of the 2990WX when rendering applications. Where it was 36% faster than the 7980XE, it is now 45% faster. It used to be 59% slower in WinRAR, now it's only 37% slower. Still slower, of course, but a much reduced margin. However, things go terribly wrong for gaming performance and now F1 2017 is completely unplayable.
Although the 7980XE scored a 1% low of 34 fps, this is a little deceptive as the game was still very playable and we only noticed a big increase in delay once or twice per round. The frame rates increased mostly by 100 fps and it was a surprisingly good experience. The 2990WX, on the other hand, was a stuttering mess and would have been impossible to play, although luckily the can benchmark took care of driving. So in this test we see a mixed performance of the 2990WX.
We are now better equipped with more information about the 2990WX's performance than a few days ago. However, this additional information has not significantly changed our perception. Throwing more work on these CPUs will only give you mixed results, while you would expect a 32-core processor to destroy an 18-core processor under these conditions.
The performance issues we've seen with the 2990WX so far may not all be due to the NUMA configuration, increased memory latency for half of the cores, or the limit per core bandwidth at full load. But there is also the possibility that the culprit will be found in the Windows scheduler. It is certainly possible that a future Windows update could improve things for the 2990WX, and we have already seen this with the first generation Threadripper CPUs.
The conclusion remains unaffected by yesterday's review. We wouldn't buy the 2990WX or 7980XE, we would buy the 2950X. The 2990WX seems risky at the moment, the mixed Windows performance and extreme price are patchy enough, but we're almost certain that AMD will offer something significantly better next year at this time with the 7nm process.
From a consumer perspective, the 2990WX is poorly timed. From AMD's perspective, we learn why they launched such a product. As always, we will continue to rate and update you if any notable changes are made.