|
[Sponsors] |
AMD Ryzen Threadripper vs Intel Xeon, importance of cache and memory channels |
|
LinkBack | Thread Tools | Search this Thread | Display Modes |
March 21, 2018, 04:15 |
AMD Ryzen Threadripper vs Intel Xeon, importance of cache and memory channels
|
#1 |
New Member
Join Date: Aug 2017
Posts: 11
Rep Power: 9 |
Hi
I am looking for a workstation computer for running CFD simulations. Typical use will be steady and unsteady simulations with 1-10 million cells, and the software will be OpenFOAM. Our budget is 4000-4500 Euro. Within the budget it seems as if our best options are either to buy a high-end gaming computer with the AMD Ryzen Threadripper 1950x and the fastest available RAM, or to buy a Workstation (we have a discount agreement with Dell, so it will be from them) with dual Intel Xeon E5-2630 v3 processors. With otherwise close to similar specs this ends up at more or less the same price. Both alternatives have a total of 16 processor cores. If evaluating the clock frequency (3.3 GHz AMD vs 2.4 GHz Intel) the AMD alternative seems to be the preferred choice. On the other hand the dual Intel alternative offers 2x20 MB "Smart Cache" and 2x4 memory channels, while the single AMD offers 16 MB L3, 4 MB L2 and 768 kB L1 cache and 4 memory channels. We have not succeeded in finding a comparison between the performance of these or similar processor alternatives on CFD cases, and my question is now whether one should weight clock frequency or the cache and number of memory channels more in this case. Any advice would be highly appreciated. |
|
March 21, 2018, 06:15 |
|
#2 |
Super Moderator
Alex
Join Date: Jun 2012
Location: Germany
Posts: 3,427
Rep Power: 49 |
Ryzen Threadripper 1950x has a total of 32MB L3, not 16MB.
That being said, a workstation with dual Xeon 2630v3 will still be faster for parallel CFD with OpenFOAM thanks to the higher memory bandwidth. TR on the other hand will be faster for single-threaded workloads. You decide which is more important for you. Intels "I9" processors or AMDs Epyc might also be an option depending on your preferences. OpenFOAM benchmarks on various hardware 4500$ might be just enough to buy the parts for a workstation with dual Epyc 7281 and 128GB of RAM. In contrast, 4500$ seems a little on the expensive side for both of the options you outlined. v3 Xeons have seen two generations of successors and you can buy refurbished workstations with them for less money. Or even a v2 which wont be much slower for much less: Dual E5-2680 V2 workstation advice I really hope that you actually get a discount from Dell and that they are not just trying to sell you their outdated hardware they want to get rid of Choosing the right option for you boils down to these factors
|
|
March 21, 2018, 08:50 |
|
#3 | |
New Member
Join Date: Aug 2017
Posts: 11
Rep Power: 9 |
Quote:
Parallel running is definitely more important than serial performance in this case, so then it seems as if your recommendation is to clearly look more at memory bandwidth than processor clock frequency. Is it so that clock frequency is close to negligible in this setting? I checked out your Epyc processor recommendation, and it seems to be possible to configure a workstation with the Epyc 7251 and 64 GB ram within our budget. My current workstation has only 32 GB ram and seldomly comes close to using all of it, so I suppose that will be sufficient. Again I see that the clock frequency is very low for this processor, but that focus is rather put on the number of memory channels. This seems to have 8 memory channels and 32 MB L3 Cache per processor. Is it then necessary to have 2x8=16 RAM chips in order to be able to exploit all the memory channels? I could not find a direct comparison of this setup with a dual Xeon setup, but is it likely that the Epyc setup will outperform the latter due to memory bandwidth? |
||
March 21, 2018, 09:40 |
|
#4 |
Super Moderator
Alex
Join Date: Jun 2012
Location: Germany
Posts: 3,427
Rep Power: 49 |
While clock speed is not a negligible factor in general, it becomes almost irrelevant as soon as you hit the memory bandwidth limit. By the way, you seem to be looking at base frequencies. Most processors operate at significantly higher frequencies even when all cores are busy. E.g. 2.7GHz all-core turbo vs. 2.1GHz base clock for Epyc 7281.
It is necessary to populate each memory channel with a DIMM in order to get any benefit. So 8 DIMMs per CPU for AMD Epyc, 4 DIMMs per CPU for Intel Xeon E5. As far as performance comparison goes, I would expect Epyc 7281 to be within 10% of Epyc 7301. More L3 cache is nice-to-have, but does not make such a huge difference in CFD. Epyc 7251 on the other hand only has 8 cores, the price/performance comparison between this CPU and the 16-core 7281 is not very favorable. Here in Germany the prices are: Epyc 7251 8-core: 465€, Epyc 7281 16-core: 575€. In my opinion this counts as a no-brainer. |
|
March 21, 2018, 13:07 |
|
#5 | |
New Member
Join Date: Aug 2017
Posts: 11
Rep Power: 9 |
Quote:
|
||
Tags |
cache, memory channels, threadripper, xeon e5 |
|
|
Similar Threads | ||||
Thread | Thread Starter | Forum | Replies | Last Post |
AMD Epyc CFD benchmarks with Ansys Fluent | flotus1 | Hardware | 55 | November 12, 2018 06:33 |
Advice on the technical requirements for a new Fluent Workstation | Dorit | Hardware | 32 | July 17, 2018 03:48 |
Insufficient RAM for Intel XEON E5-2699 v4 | alexp88 | Hardware | 4 | January 2, 2017 05:18 |
Superlinear speedup in OpenFOAM 13 | msrinath80 | OpenFOAM Running, Solving & CFD | 18 | March 3, 2015 06:36 |
Workstation with new E5 Xeon Ivy-bridge | Manuelo | Hardware | 23 | November 24, 2014 15:11 |