|
[Sponsors] |
![]() |
![]() |
#1 |
New Member
Naeem Ur Rahman
Join Date: Nov 2017
Posts: 6
Rep Power: 8 ![]() |
I'm on a budget and looking forward to build a Workstation exclusive for Ansys Mechanical, Explict Dynamics, Ansys ACP (composite modeling & Analysis) and 3d modeling using Inventor and Solidworks. [No CFD at all]
Ansys ACP usually used to much computation power due to too much details. Newer Threadripper PRO 3945wx has 8x channels memory 12x cores and excellent core clock speeds with 64 MB of L3 cache. I Will be installing 64Gb (8x8Gbs) of RAM, 512 GB M.2, 2TB HDD and some cheaper Graphics card. Is it a good build for my use or I have to go for Xeon processors |
|
![]() |
![]() |
![]() |
![]() |
#2 |
Super Moderator
Alex
Join Date: Jun 2012
Location: Germany
Posts: 3,426
Rep Power: 49 ![]() ![]() |
First things first: don't buy Threadripper Pro. Especially not the SKUs with less than 32 cores.
It is unknown if they have enough compute dies enabled to benefit from their 8-channel memory interface. Judging by the amount of L3 cache, they don't. Speaking more generally, AMD made their Threadripper Pro lineup obsolete when they launched 3rd gen Epyc CPUs. If you want to go with AMD, get a 7313p or 7443p. Intels Ice-Lake CPUs should also be a good choice. Since they have a plethora of different options, we should first find out how many cores you want/need and how much you are willing to spend on the CPU. |
|
![]() |
![]() |
![]() |
![]() |
#3 | |
Member
|
Quote:
According to the good improvement from 7542 to 7532 posted in the steaky thread, and results by https://openbenchmarking.org/test/pt...26f7a1#metrics , it seems that the memory bandwidth depends closely on the size of L3, for both 7002 and 7003. Icelake Xeon might be another story, review by Anandtech shows quite large advantage over 7003 in raw bandwith, however openbenchmarking results are not so great. I just booked one (2*6336Y) from lenovo, however it would take at least one month for manufacture and assembly... |
||
![]() |
![]() |
![]() |
![]() |
#4 |
Super Moderator
Alex
Join Date: Jun 2012
Location: Germany
Posts: 3,426
Rep Power: 49 ![]() ![]() |
128MB L3 cache on Zen3 means at least 4 compute dies active. Which is enough to saturate the read bandwidth of all 8 memory channels, same as Zen2.
The 74F3 is no doubt a great CPU. But the price increase compared to Zen3 Epyc with 128MB L3 is extremely steep. I went with a recommendation that is roughly the same price bracket as OPs original plan. The results in the link you posted seem to be all over the place. Maybe not the best resource to compare CPUs. If I recall correctly, Ice lake Xeons get their memory bandwidth advantage in the stream benchmark from hardware-level implementation of streaming stores. That's definitely a neat feature, but we have yet to see how this translates to real-world performance improvements. |
|
![]() |
![]() |
![]() |
![]() |
#5 | ||
Member
|
Quote:
However in terms of L3 reliance of bandwidth, it seems that increase from 128 to 256MB does make a difference, according to https://openbenchmarking.org/test/pt...26f7a1#metrics, that's around 30% improvement, comparing 7443 with 74F3. Currently this is the only source and so far no post of CFD benchmark on 7003, but for 7002 I think it is safe to say that bigger L3 really matters, according to 7532 (256M) of Quote:
|
|||
![]() |
![]() |
![]() |
![]() |
#6 | |
New Member
Naeem Ur Rahman
Join Date: Nov 2017
Posts: 6
Rep Power: 8 ![]() |
Quote:
1. I have $5000 for for workstation. 2. I think 7443p would be a good choice. 24 cores may be enough for me. I'm a noob to this FEM world. where to get it ordered. 3. Can I get better performance/dollar with another EPYC (Milan or Rome) or Intel Xeon CPU keeping my budget well within $5000. |
||
![]() |
![]() |
![]() |
![]() |
#7 | |
Super Moderator
Alex
Join Date: Jun 2012
Location: Germany
Posts: 3,426
Rep Power: 49 ![]() ![]() |
Quote:
From Intels latest generation of CPUs, the 6312U seems to be the closest match in terms of core count and price. As far as performance goes, it's really hard to tell at this point. Both CPUs are quite new, and I haven't seen a whole lot of relevant benchmarks for them. My personal bias would tend towards AMD. Users report it clocking close to 4GHz with all cores active. And it has the larger L3 caches. Intel on the other hand has support for AVX-512, and potentially higher memory bandwidth for codes that weren't compiled with streaming stores. I don't think you would go seriously wrong with either of these CPUs. So if one isn't available, just pick the other. |
||
![]() |
![]() |
![]() |
Tags |
3945wx, ansys explict dynamics, ryzen, threadripper, xeon |
|
|
![]() |
||||
Thread | Thread Starter | Forum | Replies | Last Post |
Xeon vs Ryzen Threadripper | Slash | Hardware | 59 | November 28, 2020 13:19 |
New workstation: Xeon, i9, Threadripper or Epyc | CRI_CFD | Hardware | 6 | January 22, 2020 10:14 |
AMD Ryzen Threadripper vs Intel Xeon, importance of cache and memory channels | JohnMartinGodo | Hardware | 4 | March 21, 2018 12:07 |
Threadripper or Dual Xeon system? | Echidna | Hardware | 1 | August 16, 2017 17:44 |
Prosurf to Proam problems, help please | William Wallace | Siemens | 11 | February 3, 2006 17:00 |