|
[Sponsors] |
Intel xeon gold vs AMD epyc build recommendations? |
|
LinkBack | Thread Tools | Search this Thread | Display Modes |
December 14, 2019, 04:47 |
Intel xeon gold vs AMD epyc build recommendations?
|
#1 |
New Member
ahmed eissa
Join Date: Mar 2019
Posts: 8
Rep Power: 7 |
i build a workstation for me in my home.
i asking for recommendations. build 1 super micro 2029TP 2u 4 x nodes X11DPT-PS (dual socket 3647) i will use 2 nodes now with an option of use all nodes in future cpu 4* intel xeon gold 6130 16 core ram 6*8gb 2666 for each cpu build 2 supermicro h11dsi (dual socket sp3) cpu 2*amd epyc 7551 32cores (watercooled) ram 8*8gb 2666 for each cpu what build is the best? thanks in advance |
|
December 15, 2019, 04:30 |
|
#2 | |
Super Moderator
Alex
Join Date: Jun 2012
Location: Germany
Posts: 3,399
Rep Power: 46 |
Quote:
Speaking of requirements: we know almost nothing about yours. Elaborate on software, licenses, budget... And usage: i.e. when you say workstation, will this sit under your desk or in a separate room. Because 1 is not really a workstation in that sense of the word. |
||
December 15, 2019, 15:43 |
|
#3 |
New Member
ahmed eissa
Join Date: Mar 2019
Posts: 8
Rep Power: 7 |
1.i do mainly external aerodynamics (no of cells about 20 to 30 million cell)
and internal flow 2.i use usually fluent and sometimes openfoam 3.the budget is 8000$ for the system and license |
|
December 15, 2019, 16:04 |
|
#4 |
Super Moderator
Alex
Join Date: Jun 2012
Location: Germany
Posts: 3,399
Rep Power: 46 |
This does not answer the question whether you want to use it as a workstation, or if you want compute nodes in a server room. Have you made up your mind about that?
Also: not really sure how you plan to fit license costs for Ansys Fluent into an 8000$ budget. This will affect the system choice. 32 or more cores are pointless, if you only have licenses for 8 of them. |
|
December 15, 2019, 16:27 |
|
#5 |
New Member
ahmed eissa
Join Date: Mar 2019
Posts: 8
Rep Power: 7 |
1.i will use it as workstation.(if the server will give me better performance per money. i will figure it out how i
put it) 2. i can increase the budget but 8k is my baseline. |
|
December 16, 2019, 02:58 |
|
#6 |
New Member
ahmed eissa
Join Date: Mar 2019
Posts: 8
Rep Power: 7 |
i have deals on both of them .
if i fill 2 nodes with cpu and ram in the server. the total price for both of the builds will be equal. but the xeon gold build (if i use 2 nodes) will have 24 memory channels. with 64 cores.(with upgrade in future .with using the 4 nodes .it will have 48 memory channels) the amd epyc build will have 16 memory channels.with 64 cores. so what will be better for me ? |
|
December 16, 2019, 05:34 |
|
#7 | |
Senior Member
Join Date: May 2012
Posts: 546
Rep Power: 15 |
Quote:
I do not fully understand. It seems that license is not an issue here, so with that in mind the question is which system is better to run your calculations on. If your question is which system will be better when comparing a 4 node 8 CPU XEON Gold system to a dual CPU AMD system then obviously the XEON system would be better. But that comparison does not seem fair in terms of investment cost... Instead of 7551 you could probably get 7301 instead with similar performance at a much lower cost (our workstations are around 4000 Euro), For 8000 $ you could perhaps get 32 memory channels worth of AMD in that case. That's what I would go for. Also, 7302 is cheaper than 7551 and would likely perform better (assuming you go with the recommended 3200 MT/s memory). The OpenFOAM benchmarks are really impressive for Rome. |
||
December 16, 2019, 12:27 |
|
#8 |
Super Moderator
Alex
Join Date: Jun 2012
Location: Germany
Posts: 3,399
Rep Power: 46 |
Agreed, these are two fundamentally different systems. Well not really systems, a lot of the configuration is still missing.
Option 1 is just a bunch of compute nodes. You might be able to get a graphics card in and a video signal out, but you probably don't want to have this in the same room with you. The noise of 1U server gear is deafening. Also, you will need a node interconnect. 10Gigabit Ethernet is probably fine for only 2 nodes, but the more nodes you add, the more performance you leave on the table without Infiniband. Option 2 can be made into a proper workstation, but it will obviously be slower than 2 or even 4 dual-socket nodes with Xeon Gold CPUs. Assuming you somehow got your hands on enough HPC licenses for Fluent. @Simbelmynė Epyc 1st gen prices are all over the place these days. Going with the 7301 no longer makes sense, because 7451 starts at 950€, and 7551 at 1200€. After you factor in the cost for the rest of the system, this is only a small price increase for 50% or 100% more cores. These are new retail prices, they are even cheaper on ebay. There is an argument to be made for the 7302, especially when per-core license costs are involved. Not sure if this is the case here I will switch my 7551 to watercooling over christmas, and see if I have any luck with memory overclocking/timings. Maybe I can claw back 1st place in the OpenFOAM benchmark. Last edited by flotus1; December 16, 2019 at 15:17. |
|
December 16, 2019, 15:00 |
|
#9 |
New Member
ahmed eissa
Join Date: Mar 2019
Posts: 8
Rep Power: 7 |
1. the company i work in will provide the license
2. i will put the server in the basement. i have 10gb network. 3. i will use 2 nodes for now so 10gb will be enough if i choose the server build . will i go to xeon gold 6138 20c or xeon gold 6130 16c i saw the open foam benchmark and i read that no gain from more cores than 16c. is this right? |
|
January 18, 2020, 23:12 |
|
#10 |
New Member
Jongtae Kim
Join Date: Mar 2009
Location: Daejeon, Republic of Korea (South)
Posts: 8
Rep Power: 17 |
As far as I know for parallel performace, number of memory channels on a board is a primary factor.
A total 16 channnel from a dual epyc board gives 16 or a little more than 16 times of speed-up at most. If you run two or more jobs with 16-cores, then the speed-up is quickly reduced below 16. 10Gb network (I think it is not a infiniband) is used then the network speed becomes another bottleneck of a parallel performance. 10Gb with 2 or more nodes with xeon is not recommended. I think the best is 2 PCs with 2-CPU epycs with lower cores connected by direct infiniband card and optical cable only when noise from epyc is controllable by water cooler. |
|
|
|
Similar Threads | ||||
Thread | Thread Starter | Forum | Replies | Last Post |
AMD Epyc CFD benchmarks with Ansys Fluent | flotus1 | Hardware | 55 | November 12, 2018 05:33 |
AMD Ryzen Threadripper vs Intel Xeon, importance of cache and memory channels | JohnMartinGodo | Hardware | 4 | March 21, 2018 12:07 |
[OpenFOAM.org] OpenFOAM build on Intel Xeon Phi | asaijo | OpenFOAM Installation | 31 | July 26, 2017 10:35 |
[OpenFOAM] Color display problem to view OpenFOAM results. | Sargam05 | ParaView | 16 | May 11, 2013 00:10 |
CFX11 + Fortran compiler ? | Mohan | CFX | 20 | March 30, 2011 18:56 |