Core Combat: Intel Xeon W-3175X vs. AMD Threadripper 2990WX

[PaulAlcorn]

AMD's Threadripper 2990WX challenges Intel's pricey Xeon W-3175X in this battle of the behemoths of the processor market.

Core Combat: Intel Xeon W-3175X vs. AMD Threadripper 2990WX : Read more

 

[sstanic]

"an unoptimized Windows scheduler"

nuff said.

 

[redgarl]

Do we really need a gaming performance segment for such a product review? Nobody is going to play on a 28-32 core CPU... especially since AMD mentioned WX were NOT aimed at gamers.

As far as features go... how Intel is winning when the only advantage it has is six-channel memory over AMD socket? 64 PCIe lanes, 64mb of L3 cache, 2933 MHz memory, 32 cores, lower temps and better power...

 

[redgarl]

"an unoptimized Windows scheduler"

nuff said.

It is way more than this. We don't even have proper benchmarks to provide a good analysis.

As of now, Phoronix is the only person who provided a good idea to what to expect of a 32 core workstation, because he was using Linux.

Microsoft needs to wake-up.

 

[Art_8_]

When you know what Windows kernel is doing to the Threadripper and you run this comparison with anything other than an attitude of tongue in cheek you are complacent in spreading a load of bull. It is well known that the scheduler was modified by MS at the request of Intel to solve an Intel processor issue and MS went ahead with it even though as a collateral outcome it crippled the Threadripper.

 

[AndrewJacksonZA]

Paul, for real, how many of these benchmarks can you run on Linux in order to mitigate the Windows scheduler issue, please?

While I am a Windows user and all of my & my team's current use cases all fall on Windows software except for two which can run on Linux, but don't (yet?) there are many people who are not in that situation.

Also, seriously, gaming benchmarks??

 

[A Stoner]

Just because you are not going to buy a 28 or 32 core processor for playing games does not in and of itself mean that gaming performance is not important.
If I were going to dump $8000 to $12000 on a computer for rendering for example, I would not want to have to turn around and drop another $3000 to $4000 on a gaming computer.

 

[mischon123]

Love to have a Threadripper. But right now life is good. 8 Ryzen cores already make this the best computing platform ever. Everything zips. No delays, stoppages, no weird bugs. Just happy.
Those chips show whats ahead once they solve the bottlenecks, integrate ram and better graphics near the cores. I guess we will have 128 cores/128mb Ram/128gb graphics pipeline in a while. 2,4,6,16,24/32/64/128 in 2025?

 

[PaulAlcorn]

Paul, for real, how many of these benchmarks can you run on Linux in order to mitigate the Windows scheduler issue, please?

While I am a Windows user and all of my & my team's current use cases all fall on Windows software except for two which can run on Linux, but don't (yet?) there are many people who are not in that situation.

Also, seriously, gaming benchmarks??

Running Linux isn't an issue, but spinning up benchmarks for multiple chips is a time consuming endeavor. Windows is the predominant OS for workstations, whereas we use Linux for server testing.

Yes, we run games. We do not deviate from our standard test regimen on any chip. This is because some might think we are hiding poor performance, etc, for the device under test. However, we will add tests to reflect specialized use cases, such as the workstation tests above.

 

[tyns78]

Performance per dollar is a different category from performance; to claim AMD ties or beats Intel in some categories b/c it's cheaper is disingenuous. Have performance comparisons, then have value comparisons.

Intel smokes AMD's product, period. You'll always have to pay much more for the best hardware.

 

[WINTERLORD]

a amd motherboard is way cheaper then the MB that fit the intel chip plus cost of chip amdd wins this assuming the performance is close to same. although for just an ordanary high end machine iv always loved intel however they got some serios compatetion here

 

[InvalidError]

"an unoptimized Windows scheduler"

nuff said.

It is way more than this. We don't even have proper benchmarks to provide a good analysis.

For massively parallel systems, you need system architecture specific optimization to achieve good scaling or you end up with threads thrashing each other all the time and heavily sub-optimal result. TR gets hit pretty bad there for being a NUMA beast and most software not being optimized for anything like that.

 

[AndrewJacksonZA]

Thanks Paul.

 

[10tacle]

So AMD wins on price, Intel wins on overall performance, and yes, that INCLUDES gaming. I can see this processor at the core of a professional flight simulator or racing simulator setup. For example, if anyone has ever been to an Andretti Racing sports/gaming place (US only as far as I know), check out their CXC Motion Pro II driving simulators that are controlled by one PC running SAP as the track, car, and driver management software. They have six simulators tied together with each having three 50" monitors attached to fully articulating motion seats.

 

[dario.diablos]

Being a professional tech journalist, you must certainly be aware of the windows scheduler fix available for Threadripper. Does it not then seem disingenuous to test without utilizing that fix? It's a free bit of software, after all, and one which the enthusiast community is well aware of....or perhaps I was being generous with the word "professional".

 

[emeralds1000000]

Hello Tomshardware ,

your new comparison Conclusion table is not professional at all. you cant compare products lke this and put an X then count the x and say Tie.


Because each field does not have the same weight. This needs Weight Function .

an X in cooling for example can never equals an x in rendering at all. this is kiddo comparison . you give them the same weight ?

Remove this unprofessional way ASAP from Tomshardware

 

[mdarrish]

That is 56, not 58 threads. Proofread!

 

[linkendorf]

Um..28x2=56. Typo in the second paragraph.

 

[PaulAlcorn]

Being a professional tech journalist, you must certainly be aware of the windows scheduler fix available for Threadripper. Does it not then seem disingenuous to test without utilizing that fix? It's a free bit of software, after all, and one which the enthusiast community is well aware of....or perhaps I was being generous with the word "professional".

CorePrio, which is the core component of the Dynamic Local Mode that AMD enables in the software mentioned in the article, and the new NUMA Dissociater feature within that program. Yes, we're aware of the work that is being done on that front, much of it by Wendell from Level1 Techs.

However, there is currently some performance regressions in areas, at least as of the last time we checked up on that project (around the NDA lift for this article). We'll be happy to test with it as the project progresses to a more mature state. For now, it hasn't been validated with the 2990WX by the team working on the project, as noted in the documentation:

The NUMA Dissociater is confirmed to work on EPYC 7551 and TR 2970/2990, but may work on other HCC NUMA platforms.

This is very much a work in progress by members of the enthusiast community in an unsanctioned effort, and there are still several unknowns:

The ‘fix’ is also bizarrely imprecise. For affected processes, a call to SetProcessAffinityMask, without even changing the affinity (e.g. all CPUs to all CPUs), resolves it – at least most of the time. Best guess is that the preferred NUMA node for the process is removed and that causes the Windows scheduler to change behavior, as evidenced by the thread ideal processor selections, and more importantly the massive change in performance.

+Just for reference, this is not a 'fix' for the Windows scheduler, either. It does not make any alterations to the scheduler whatsoever.

It has been difficult to reach solid conclusions because the behavior is so bizarre. Hopefully the community can contribute to our understanding of the details and scope of this sub-optimal behavior.

Update: It must be noted that te NUMA Dissociater doesn’t ‘take’ sometimes, particularly on the 2990wx. EPYC is less impacted. On the 2990wx, restarting the Indigo process 10 times, 50% of those instances won’t result in the ‘fast’ edition. This is still a massive improvement, as without the NUMA Dissociater, 100% of runs will be slow. This is not any bug in Coreprio, but something inherent to the nature of the performance issue.

We're not afraid of beta testing things, but given time constraints, it's best to wait until this project is further along.

It should go without saying that these types of unexplained and erratic behaviors aren't going to lend themselves well to the target workstation market.

 

[PaulAlcorn]

That is 56, not 58 threads. Proofread!

Thanks for the heads up, fixed!

 

[PaulAlcorn]

Hello Tomshardware ,

your new comparison Conclusion table is not professional at all. you cant compare products lke this and put an X then count the x and say Tie.


Because each field does not have the same weight. This needs Weight Function .

an X in cooling for example can never equals an x in rendering at all. this is kiddo comparison . you give them the same weight ?

Remove this unprofessional way ASAP from Tomshardware

I agree that weighting is the best approach, but unfortunately deriving a weighting metric would involve opinion. How much weight do we assign to each category?

This would lead to immediate criticism, such as:

Is that unfairly skewing results?

Does Tom's Hardware value single-threaded over multi-threaded, etc etc?

Do we not assign enough weight to value, or conversely, raw performance?

Then we would have to come up with different weighting mechanisms for each class of processor, as different processors are designed for different use-cases. That opens yet another can of worms, and criticism.

We eschew weighting in favor of presenting raw data, which in turn allows the reader to select areas of interest for them, and then make decisions accordingly. Simply put, if a certain category isn't important to you, remove that checkmark from your calculation.

 

[z3.lionking]

<Quote>Do we really need a gaming performance segment for such a product review? Nobody is going to play on a 28-32 core CPU... especially since AMD mentioned WX were NOT aimed at gamers.

As far as features go... how Intel is winning when the only advantage it has is six-channel memory over AMD socket? 64 PCIe lanes, 64mb of L3 cache, 2933 MHz memory, 32 cores, lower temps and better power...</Quote>

Please take this review with a bucket of salt! )

 

[z3.lionking]

Just because you are not going to buy a 28 or 32 core processor for playing games does not in and of itself mean that gaming performance is not important.
If I were going to dump $8000 to $12000 on a computer for rendering for example, I would not want to have to turn around and drop another $3000 to $4000 on a gaming computer.

-------------

You need to understand that the amount of content to be "rendered" that an individual can create.. would easily be handled by a Ryzen 7 or Core i7.

Don't just throw the word "rendering" around senselessly.

Do you know how much time would it take to create content that probably would take 1 hour to render on a Ryzen 7 2700X? At the very very minimum - 1 full day!

So no individual would throw away $8000 to just save maybe 30 mins on a project thats anyway going to take days to complete!!!

The gaming benchmarks are extremely stupid and there is no sensible justification for it. (Benchmarks for the sake of benchmarks implies you are stupid - which is whats happening here.)

 

[redgarl]

Paul, for real, how many of these benchmarks can you run on Linux in order to mitigate the Windows scheduler issue, please?

While I am a Windows user and all of my & my team's current use cases all fall on Windows software except for two which can run on Linux, but don't (yet?) there are many people who are not in that situation.

Also, seriously, gaming benchmarks??

Running Linux isn't an issue, but spinning up benchmarks for multiple chips is a time consuming endeavor. Windows is the predominant OS for workstations, whereas we use Linux for server testing.

Yes, we run games. We do not deviate from our standard test regimen on any chip. This is because some might think we are hiding poor performance, etc, for the device under test. However, we will add tests to reflect specialized use cases, such as the workstation tests above.

Nobody buying those CONSIDER gaming in their purchase... so yes, it is misleading.

 

[redgarl]

<Quote>Do we really need a gaming performance segment for such a product review? Nobody is going to play on a 28-32 core CPU... especially since AMD mentioned WX were NOT aimed at gamers.

As far as features go... how Intel is winning when the only advantage it has is six-channel memory over AMD socket? 64 PCIe lanes, 64mb of L3 cache, 2933 MHz memory, 32 cores, lower temps and better power...</Quote>

Please take this review with a bucket of salt! )

ROFL, my comment was deleted. because it was asking the right questions.

It is not really the products that pissed me off, it is the biased process that are totally arbitrary.

In the end, you are comparing a product that was not design for some aspects, by using benchmarks not proving anything because the OS is gimping the result on one of the platform, and that cost 3 times less than the other.

I mean sure, if you want, I am pissed about the fact that Intel tied when it should have lost... in a biased comparison with questionable choices of tests... with questionable review process with an author who is not able to take the heat by confronting everyone not agreeing with his methodology and his conclusion.

Basically, as an electrical engineer having to review process in projects and products, if someone was doing something like that in my department, I can assure you that a chat would be in order.