[Habib-CFD]

The i5-12600k looks like a new flagship for the mainstream CFD Platform.


Link


Picture1.jpg

[casseer15]

Originally Posted by Habib-CFD

The i5-12600k looks like a new flagship for the mainstream CFD Platform.


Link


Attachment 87098

But these still have dual channel memory, right?
would you recommend these for CFD then?

[flotus1]

Intels new 12th gen Alder lake desktop CPUs are pretty fast. They offer just about the highest single-core performance of any CPU today. And thanks to the high bandwidth of DDR5 memory, they are also the fastest desktop CPUs for CFD overall.
The issue is that the early adopter fee blurs the line between this mainstream platform, and established HEDT platforms with DDR4. Your are looking at 300€ for an I5 CPU, at least 200€ for the cheapest DDR5 capable motherboard, and currently around 7€/GB for the cheapest DDR5.

[Cinek_Poland]

Which set should be faster in Abaqus/Ansys using 14 or 16 cores ?

1) Intel 12900k + 128GB dd4 (2x64gb)
2) Epyc Milan 7313 + 128gb ddr (8x16)

?

[flotus1]

A few things worth noting there:
1) I9-12900k only has 8 performance cores. The other 8 cores are slower "efficiency" cores. That's one of the defining features of this CPU generation. For anything CFD-related, using different types of cores simultaneously comes with a lot of caveats and drawbacks. Quite frankly, I would only count the performance cores. Trying to utilize the the efficiency cores simultaneously just is not worth the hassle.
2) There are no compatible 64GB DDR4 memory modules for Alder Lake. Or any other mainstream desktop CPU for that matter. 32GB is the current maximum for UDIMM. There will be 64GB DDR5 UDIMM modules sooner or later, but not yet.
3) Using DDR4 memory with Alder Lake is throwing away most of the benefit of this new and expensive platform. It's either DDR5, or a different -overall cheaper- platform.
4) A single Epyc 7313p is the better CPU for CFD. Especially if you can use 16 cores.

[aparangement]

Alderlake with ddr5 should be much faster than tigerlake. But this benchmark results looks quite strange, 3960 is even slower than 5950, is this for single thread only?

Originally Posted by Habib-CFD

The i5-12600k looks like a new flagship for the mainstream CFD Platform.


Link


Attachment 87098

[flotus1]

Even if it was just single-threaded... the Euler 3D "CFD" benchmark is all over the place. I have more trust in a dice-roll to determine the performance of a CPU, than this benchmark.

[Cinek_Poland]

https://www.igorslab.de/en/intel-mac...-fuer-amd-2/8/

what do you say about it ? 12900k with 32gb ram is about 42% better than 5950x , I was hopping that 12900k would be very close epyc 7313 in FEM not exactly CFD.
Do you think 12900k + 128gb platform (2x64 ddr5 in future) would be close (in performance ofcourse) epyc 7313 +128gb (8x16) in FEM (mechanics like plastic deformation not CFD) ?

Why ddr5 for this processor would be better than ddr4 ? I dont see much difference

[Habib-CFD]

Originally Posted by aparangement

Alderlake with ddr5 should be much faster than tigerlake. But this benchmark results looks quite strange, 3960 is even slower than 5950, is this for single thread only?

I agree with you, looks strange! based on the details the benchmarks run at the maximum available threads.

[flotus1]

Originally Posted by Cinek_Poland

what do you say about it ? 12900k with 32gb ram is about 42% better than 5950x

And most of that lead comes from the higher transfer rate of DDR5

Originally Posted by Cinek_Poland

I was hopping that 12900k would be very close epyc 7313 in FEM not exactly CFD. Do you think 12900k + 128gb platform (2x64 ddr5 in future) would be close (in performance ofcourse) epyc 7313 +128gb (8x16) in FEM (mechanics like plastic deformation not CFD) ?

Unless you are running some weird-ass FEM solver that is mostly single-core limited, not a chance.

Originally Posted by Cinek_Poland

Why ddr5 for this processor would be better than ddr4 ? I dont see much difference

Which specs or results are you looking at here? DDR5-4800 is straight up 50% higher transfer rate than DDR4-3200. Sure, that value is rather theoretical, and the test you linked uses DDR5-5200 vs DDR4-3777. But it shows in the benchmarks that are sensitive to memory bandwidth. It just makes no sense to run these CPUs with DDR4, especially for FEM or CFD.

[aparangement]

How about the 503 subscore of SPEC2017?

Its website says it is a CFD benckmark that "numerically simulates blast waves in three dimensional transonic transient laminar viscous flow"

https://www.spec.org/cpu2017/Docs/be....bwaves_r.html

And the scores looks reasonable. e.g.
https://www.anandtech.com/show/17047...d-complexity/8

and
https://www.anandtech.com/show/16594...lable-review/6

Originally Posted by flotus1

Even if it was just single-threaded... the Euler 3D "CFD" benchmark is all over the place. I have more trust in a dice-roll to determine the performance of a CPU, than this benchmark.

[Iuvatix]

Hello everbody,


has anybody testet an Alderlake CPU with the OpenFoam testcase or other RANS CFD codes/software and compared to older Intel and Ryzen CPUs?
I think the increased single thread speed and DDR5 could make these CPUs very good allrounder for many applications.

[flotus1]

It's not a huge secret that Alder Lake desktop is hands down the fastest mainstream CPU right now. Especially for CFD. That much I can tell you without any CFD-specific benchmarks.
The only problem is the early adopters tax. DDR5 memory is expensive. Even more so if you want something faster than DDR5-4800. Motherboards also had quite the price jump. It's not so bad anymore now that B650 and H670 are out. But I still can't shake the feeling that the supply/demand issues we had over the last 2 years for some components, have had a lasting impact on prices for other parts too.
If your budget allows it, you can't go wrong with these CPUs. Provided that mainstream CPUs fit your requirements of course.

[Ford Prefect]

According to this site, the 5900X and 5950X is equally fast as the 12900K for the 30M OpenFOAM benchmark case. The site publishes a bit strange results though, so it should be taken with a grain of salt. I think that Alder Lake with good DDR5 memory would be superior to Zen 3. There is a lot of really poor DDR5 memory out there now. I mean in terms of latency. Even with the higher frequencies considered they have much worse latency than good DDR4.


I think 12900k might also have too high power output to cool for an extended period of time. Another issue is the hybrid design if you are on Linux. Right now the state of the scheduler is becoming better, but it is not perfect and sometimes efficiency cores get jobs ahead of performance cores when they shouldn't. I would probably disable the e-cores and test performance.



https://openbenchmarking.org/test/pts/openfoam

[LastAirBender]

Hi, would you suggest using e-cores for CFD as well? I plan to limit all cores of an i9 13900hx (8+16) to 3.2ghz and run it for long durations. Does this seem feasible to you in your experience? or should I stick with i713700hx which has the same number of p cores? I am planning to buy a laptop, I can't buy a desktop due to the mobile nature of my work.
Regards

[Jajaw]

Hello,

Looking at local universities is indeed a very good idea. I was able to get a box that is still in very good condition for a lower price with this method in the past.