The Mega-Tasking Test: AMD Threadripper 2990WX Heavy Multitasking Benchmark

[Julio Franco]

After our long look at the new Threadripper 2990WX and 2950X processors, many called us out on benchmarks that only ran a single application at a time. We had already planned to make a separate multi-tasking focused benchmark, and so here we are after a few additional days of testing with more results to discuss.

Read the full article here.

https://www.techspot.com/review/1680-threadripper-2-mega-tasking/

 

[Sausagemeat]

Interesting to see how chips perform in games whilst encoding or rendering. Certainly you’d rather be able to play games whilst leaving something encoding in the background. Especially if you’re making gaming videos or so. Interestingly I’ve been watching Mumbo jumbo before and seen his frame rate tank and for him to go “oh it’s because I’m rendering in the background”.

 

[ET3D]

I think that the 2990WX is a good product for those who need to use its particular strengths. Or those using Linux. I think Phoronix was quite convincing that it's Windows that's the problem.

 

[Maxiking]

So, it is a 32 core cpu terrible at multitasking ???

 

[Charlie Varburg]

When you are a 3d artist, an engineering, an architect etc there are specific workloads and specific applications you are running at the same time. The list can go on and on.

This article was nice idea but is poorly implemented. All the applications that have been used are not real-life scenarios for a professional workflow. It is very clear that the reviewer has not used any scientific approach or have the necessary academic knowledge to simulate that workflows properly.

It is not easy test all workflows but we are talking for a CPU that is targeted to specific scenarios and it should be tested like this to make safe conclusions. I hope AMD in the next version to outcome all these problems and have better cooperation with the software houses for having optimized applications for their processors.

 

[Shumbo]

When you are a 3d artist, an engineering, an architect etc there are specific workloads and specific applications you are running at the same time. The list can go on and on.

This article was nice idea but is poorly implemented. All the applications that have been used are not real-life scenarios for a professional workflow. It is very clear that the reviewer has not used any scientific approach or have the necessary academic knowledge to simulate that workflows properly.

It is not easy test all workflows but we are talking for a CPU that is targeted to specific scenarios and it should be tested like this to make safe conclusions. I hope AMD in the next version to outcome all these problems and have better cooperation with the software houses for having optimized applications for their processors.

I saw the Phoronix testing and it sure made the 2990WX look good. I wonder if this is really a windows/linux difference or more application specific?

 

[Shknawe]

These tests are like asking an 8 armed man to only use one arm to put a bike together. Get some professionals to use it in their daily work and see what the results are. Then let the chips fall where they may. No pun intended.

 

[Julio Franco]

I think that the 2990WX is a good product for those who need to use its particular strengths. Or those using Linux. I think Phoronix was quite convincing that it's Windows that's the problem.

I saw the Phoronix testing and it sure made the 2990WX look good. I wonder if this is really a windows/linux difference or more application specific?

I'll just copy what Steve commented in our original Threadripper 2 review:

Yeah, there is a performance uplift for the 2990WX, but guess what, there's also a performance uplift for the Intel CPUs and you can see that by comparing the two Phoronix articles, they probably should have included the 7980XE alongside the 2990WX. That would have helped with the confusion

Pov-Ray: Our results on Windows: 36.3% lower completion time for 2990WX, Linux 34.2%.

Blender: Using Windows, we found the 2990WX to complete Blender in 20.3% and 30.8% less time than the 7980XE. Phoronix found on Linux a margin of 40.2%, 32.1% and 26.8%, in different tests of course.

So yes the 2990WX is faster using Linux, but the Intel CPUs are equally faster

Ohh 7-zip is the only odd result but I'd say it's a Windows bug.

 

[Foobario]

Poorly timed if you are a Windows user, perhaps.

Fortunately, the majority of high, high end users left Windows years ago.

In the original review it seemed a bit disingenuous that you failed to realize that the 2950X had already been optimized for Windows since it's onetime unique configuration mimics the 1950X. Which Microsoft only optimized after initial reviews had been done last year, as well.

If I was a conspiracy nut I might think Microsoft's inability to to provide timely optimizations might be due to allegiance to the other half of the X86 duopoly.

At the end of the day you might be right in regards to the conclusion that the 2950X is the sweet spot. As it seems there are still diminishing returns as core counts go higher. Even when the Intel 22 and 28 core chips debut the ridiculous prices they will carry will negate any superiority in performance they might bring. Although, with six channel memory they may destroy all Threadrippers in the memory intensive benchmarks.

 

[Shumbo]

I'll just copy what Steve commented in our original Threadripper 2 review:

Yeah, there is a performance uplift for the 2990WX, but guess what, there's also a performance uplift for the Intel CPUs and you can see that by comparing the two Phoronix articles, they probably should have included the 7980XE alongside the 2990WX. That would have helped with the confusion

Pov-Ray: Our results on Windows: 36.3% lower completion time for 2990WX, Linux 34.2%.

Blender: Using Windows, we found the 2990WX to complete Blender in 20.3% and 30.8% less time than the 7980XE. Phoronix found on Linux a margin of 40.2%, 32.1% and 26.8%, in different tests of course.

So yes the 2990WX is faster using Linux, but the Intel CPUs are equally faster

Ohh 7-zip is the only odd result but I'd say it's a Windows bug.

Did we see the same Phoronix review? In the review I saw, the 2990WX dominated the benchmarks. Maybe lost 1 single benchmark out of what? 20? I didn't get an exact count...but must have been over 90% won by significant margins. Many different benchmarks done...not just Pov-Ray and Blender. I don't know enough to say how important these different applications are, but no one would get the same impression of the 2990WX when comparing the Techspot and Phoronix reviews.

 

[xylreen]

So, it is a 32 core cpu terrible at multitasking ???

you should do "handbraking" and then "zipping" while "baking" your self and compare your result with author's result

 

[xylreen]

When you are a 3d artist, an engineering, an architect etc there are specific workloads and specific applications you are running at the same time. The list can go on and on.

This article was nice idea but is poorly implemented. All the applications that have been used are not real-life scenarios for a professional workflow. It is very clear that the reviewer has not used any scientific approach or have the necessary academic knowledge to simulate that workflows properly.

It is not easy test all workflows but we are talking for a CPU that is targeted to specific scenarios and it should be tested like this to make safe conclusions. I hope AMD in the next version to outcome all these problems and have better cooperation with the software houses for having optimized applications for their processors.

ditto, though it's more than you suspect, from marketing view perspective the 32cores could just be AMD's strategy to torn down intel's decade of flagship CPU, and now prosumers become aware that AMD's products are not 2nd class

 

[ET3D]

At the end of the day you might be right in regards to the conclusion that the 2950X is the sweet spot. As it seems there are still diminishing returns as core counts go higher.

That's always true for CPUs. The highest end CPUs are usually not cost effective as lower end one. True that in this case the 2990WX is also not a great choice for a do-it-all PC, but for those who want the best performance in the particular applications that benefit from it the 2990WX should be the best thing to get.

 

[dennist54]

Nice article. The article would be more interesting if the 2950X was included in the comparison. I am interested to see how the 16 core cpu does with the Stockfish nps.

 

[vexcro]

It is normal that so much cores are slower in some applications. Xeons with more than 20 cores are slowe in After Effects than Xeons with under 20 cores. And I think that 64 gigs of ram isnt much for 32 cores...

 

[motogplvr]

I own a 1950x and encode to hevc regularly. I would suggest to people reviewing a 32 core CPU to run two simultaneous instances of x265 instead of running a single instance of x265 via handbrake. This isn't going to saturate a 32 core CPU even encoding from 4K to 4K. I honestly look at reviews of these CPUs online and it pains my head to read the lack of thought into really getting an accurate measure of performance. Install the hyper-v role and create a couple VMs with 16 virtual CPUs (or 32 each) and see what performance you get encoding 4K to 4K x265. For ease of use you can just use ripbot264 that has built in distributed encoding. I see these reviews people encoding into 1080p or 4K with a single instance and it's just amateur frankly. Think about actual use cases for people buying a 32 core CPU not people encoding a lame video of themselves at da club into 1080p for their iphone. I'm going to take a wild guess than you are using far less than 50% of the CPU with your testing.

 

[motogplvr]

I own a 1950x and encode to hevc regularly. I would suggest to people reviewing a 32 core CPU to run two simultaneous instances of x265 instead of running a single instance of x265 via handbrake. This isn't going to saturate a 32 core CPU even encoding from 4K to 4K. I honestly look at reviews of these CPUs online and it pains my head to read the lack of thought into really getting an accurate measure of performance. Install the hyper-v role and create a couple VMs with 16 virtual CPUs (or 32 each) and see what performance you get encoding 4K to 4K x265. For ease of use you can just use ripbot264 that has built in distributed encoding. I see these reviews people encoding into 1080p or 4K with a single instance and it's just amateur frankly. Think about actual use cases for people buying a 32 core CPU not people encoding a lame video of themselves at da club into 1080p for their iphone. I'm going to take a wild guess than you are using far less than 50% of the CPU with your testing.

You can also try 4 VMs running 16 virtual CPUs if 2 VMs does not saturate the CPU. Really a review of a 32 core CPU w/o running virtualization is not professional. Professionals will be using virtualization with one of these CPUs to split their workloads.