[ad_1]
To present a full article for launch, my look at AMD's 2990WX and 2950X Ryzen Threadripper combined Windows and Linux performance in the same article. It turned out to be a mistake, since few people noticed had Linux tests, despite an obvious demand for them.
Before publication, I debated whether or not I should divide Linux performance into its own article, but in this particular case, I opted for the combo because I thought the most big picture was needed. This is because in Windows, scaling performance on such a large processor is a failure, while the Linux kernel seems to support the biggest problem of AMD.
I'm not going to sit there (or sit down) and pretend to understand why the 2990WX does not work as well in all Windows tests, because it's hard to get a clear answer from anyone. Nobody wants to get around the blame, but obviously, Windows seems to be the main problem. This article exists not only to draw attention to this, but also to highlight a little better what the 2990WX is capable of – if the software in question can take advantage of it.
Instead of copying and pasting the previous Linux results in this new article, I have re-tested the Core i9-7980XE 2990WX and Intel with a newer kernel (4.18.5, vs. 4.15.0) and added some additional tests. If you want to see more in-depth results for six other processors, I recommend you review the 2990WX Launch Article.
Almost all of the Linux benchmarking I do is done through the Phoronix test suite, which makes it almost also easy to generate a lot of useful results very quickly (well, depending on the hardware, of course). Unlike the introductory article, this one has 13 results in total (out of 8), which represents some extra angles where a 64-thread processor can shine if it has the appropriate opportunity. .
Before testing, I run this command as sudo to apply the performance profile:
echo performance | tee / sys / peripherals / system / cpu / cpu * / cpufreq / scaling_governor
As I mentioned in the 2990WX launch article, running this command on the 2990WX machine returned an error stating that the file did not exist. Interestingly (or is it?), This error does not exist anymore on the new kernel, so that's a plus. At the same time, the performance did not change much between 4.18.5 and 4.15, so the error did not necessarily mean too much.
Compile
I admit that before testing, I was expecting the 2990WX to encroach on the top of the Intel stack, but that does not happen. With the ImageMagick compilation, Intel is almost perfectly associated with the AMD chip that has almost twice as many cores. The kernel compilation test is much better for the 2990WX, but still, I think the performance could even better with the right software.
Blender
One of the most interesting uses of high performance hardware is rendering, and believe me, rendering engines do not fail to take advantage of every heart you can offer them (processor or graphics processor). With our Blender test, the Cycles rendering engine is used on only the processor, and when it does, AMD's many cores allow it to surpass Intel's high-end desktop chip – and quite easily.
Blender is very well optimized for these many basic workloads, and seeing this performance makes me even more enthusiastic about version 2.80 of the suite, as it will introduce heterogeneous rendering capabilities (CPU + GPU). Based on our previous V-Ray tests, this design could result in a dramatic improvement in rendering performance.
Hand brake
Despite the redundancy, I tested HandBrake in both Linux and Windows for the launch article, as it's always interesting (for me) to see performance differences for a single test on two operating systems different. Going down that path turned out to be a blessing, because while I was using a Linux version, I was using another, more recent one in Windows – and this version had big problems with Zen.
This version is 1.1.1, which remains the current stable version available on the official website of HandBrake. I used 1.1.1 for Windows testing, which resulted in Zen-based chips taking twice as long for each encoding, while Linux version 1.1.0 was no problem, and we has given a decent evolution. In fact, even the new versions of HandBrake in the tests offer more performance is increasing, not just for Zen.
For x264, the performance between 1.1.0 and the night is the same for AMD and Intel. The x265, however, shakes things up a bit, bringing improvements to both chips, even though Intel does not earn as much as AMD. The difference is simply amazing, needless to say: if you are a HandBrake user who benefits from H.265, you will want to recover the last night.
Laser plot
Rendering and ray tracing are two peas in a pod, so it is not surprising to see a large scaling between the two processors in these tests. I must note that none of these ray tracers are considered "current", but some have been reiterated in the new rendering engines. This does not mean that the performance of these tests is useless, because like all the good renders in bytes, the ray tracers are designed to evolve, and these three tests do it very well.
It's hard to ignore the fact that AMD is just stammering Intel into the little test, which is not unusual since Zen was abandoned last spring. I do not know how this particular test reflects performance in the current landscape, but scaled security extends to Tachyon or C-ray, as Blender has shown, a perfect synthetic scaling. is not always perfect However, the ability to scale is clear here.
Scientist
Scaling is the name of the game with scientific tests, so this set Rodinia takes advantage of the 64 threads of the 2990WX. Scaling is better with the LavaMD test than the solver, but a 33% performance gain at the low end is hardly difficult to evaluate.
encryption
AMD is excellent in cryptography, and this is easily verified here. That said, this is a test that Intel typically performs like a hot knife in butter. However, the pressure performance offered by AMD places it at the top of the list of winners. With such a result, it is clear that Intel could be extremely competitive if it decided to cross its barrier of 18 cores on the desktop (we are still waiting for the 28-core posted at Computex).
Compression
This set of results could be the most interesting in the article. In the launch review, the 2990WX beat the Intel chip of 2,000 MIPS, but with the kernel upgrade, AMD's performance has fall. I had to check the health of it, and it stuck. For some reason, 4.15 provides better performance for this test. This is not a good reason to use an older kernel, and it may be that the next kernel will fix it again. This is not the last time I'm testing the 2990WX, so I'll test once again versions 4.20 (or 5.0, if it's skipped).
For reasons of interest, the same compression test on Windows only reaches 55,000 MIPS on the 2990WX. Even with the drop in performance, it is still far from the performance of this OS.
Memory
This result requires a little explanation, because it is just to convey it as definitive. AMD beats Intel by about 10%, which is noticeable, but it is also because the AVX-512 has not been introduced. On Windows, Sandra, from SiSoftware, will use the most appropriate instruction set. Thus, some processors and tests may be useful. Only the high end of the Intel battery has this, therefore, for a more complete look, I recommend you to launch the launch article.
These results also do not show any problem with the memory design of the 2990WX. The overall bandwidth is correct, but it is the latency that makes the chip less ideal for non-intensive purposes – like games. Due to the weather, I have not explored the memory on the 2990WX as much as I would like, and it may take a little while before I can access it because of other content that needs special attention. In the end, the 2990WX offers good bandwidth, but this result only represents part of the picture. That's why it's really interesting to know your workload: you should not use a 2990WX on a chip like the 2950X or 7960X / 7980XE unless you know your workload.
Additional tests
For this set of results, I decided to opt for a relative performance because it was easier to keep them all on the same chart. In the case of Darktable and Hackbench, the results are expressed in seconds and the lower values are better. The others have distinct values, but are always higher, the better.
With Darktable, which is a clone of Adobe Lightroom, and IndigoBench, a rendering engine, the overall gains are minor, but still remarkable. With Darktable, Intel's multimedia prowess puts it very close to AMD's much larger chip. Things change dramatically with the other tests, however.
With the HPC challenge, the raw computing power of the 2990WX allows it to earn 55% compared to the Intel chip. With HackBench, a reference for system calls, AMD is working extremely well again. Even the 44% gain in the chess engine test is impressive. I do not really see a lot of chess fans incorporating a 2990WX in their game, but how great is it to be able to enjoy it?
Last thoughts
As you can probably see, AMD's 32-core, 32-core Ryzen Threadripper 2990WX can boost some of the tests and impress others. You may have noticed a lack of single-precision testing, and that's because I was focusing on multithreaded tests to show what happens when things get complicated. It should be no secret that single-threaded performance on a 32-core processor will not be near the best in the market (single-threaded Windows tests have been released in the launch article). This has no impact on regular use, but the fact is that you will experience lower performance in scenarios that rely heavily on single-thread performance.
This is the perfect example of a product that emphasizes the importance of knowing your workload. Some applications do not fit well, even if they seem to do it. I did not have this problem in Linux which I remember, but under Windows with some tests, even if 100% of the processor was used, the gain on half of the threads was not much different.
With HandBrake here under Linux, not all threads will be used, so the 64-core chip does not show its strength as well as possible. To work around this problem, I could have run the FFMPEG test that generates an FPS result, but I'm not sure how relevant this benchmark is to the real world because I could not see the same type of scale elsewhere. The same could be said about Darktable; the benchmark will use the entire processor, but I have never been able to get more than half of the 2990WX threads used in actual tests (but that does not mean it's not possible;
If this article does not cover the type of performance you were looking for, please leave a comment and I'll see if I can fix it the next time the machine is connected (the 7980XE and 2990WX are in their own dedicated PCs, so they're usually practice). You can also check out the 2990WX launch article, even though it focuses primarily on Windows because the overall performance can still be transferred to your Tux solution.
There are more 2990WX tests to come, but other content and launches will allow me not to talk too much about it in the near future, but the suggestions of relevant tests are appreciated, as well as any feedback in general .
Source link
