CFD Online Logo CFD Online URL
www.cfd-online.com
[Sponsors]
Home > Forums > General Forums > Hardware

How to balance the CPU and the memory bandwidth on a HEDT platform?

Register Blogs Community New Posts Updated Threads Search

Like Tree1Likes
  • 1 Post By flotus1

Reply
 
LinkBack Thread Tools Search this Thread Display Modes
Old   March 11, 2019, 20:36
Default How to balance the CPU and the memory bandwidth on a HEDT platform?
  #1
New Member
 
Hao Tang
Join Date: Feb 2019
Posts: 3
Rep Power: 7
mousefire is on a distinguished road
As an unprofessional CFD user, I'd like to establish a computer system fit for both CFD,FEM calculation and normal use (even something such as lightroom, which also needs high per-core performance). Also the budget is limited to about 3000 ~ 4000 dollars, so I suppose an HEDT system be suitable. My CFD cases usually run on fluent and CFX, up to 10M~20M grid numbers.

Still some problems:
1. How to balance the CPU and memory bandwidth? X299 and X399 platform both have 4-channel memory support and I learned from the forum that CFD calculation is sensitive with memory bandwidth. Will it be helpful if I use 8*DDR4 3200 memory configuration(also want to know the stability of the over clocked memory when running CFD calculations)? And how many CPU cores won't be throttled by such memory configuration? 12 or just 8?Is there a proper algotithm of balance between CPU and the memory band width?

2. Nowadays the intel CPUs offer better AVX floating-points calculation performance and fluent uses avx2 techonology. So will an intel platform be better? or even wait for the AMD zen2 which claims to have 2*avx performance compared to current zen cpus.

Suggestions are appreciated!
mousefire is offline   Reply With Quote

Old   March 12, 2019, 15:19
Default
  #2
Super Moderator
 
flotus1's Avatar
 
Alex
Join Date: Jun 2012
Location: Germany
Posts: 3,399
Rep Power: 46
flotus1 has a spectacular aura aboutflotus1 has a spectacular aura about
Seems like you need a bit of everything: parallel performance, single-threaded performance and memory bandwidth. So there has to be a trade-off.
For Lightroom (and many other applications): https://www.pugetsystems.com/labs/ar...l-vs-AMD-1298/
The general rule of thumb for CFD: 2-3 cores per memory channel. More if you don't have to pay a per-core licensing scheme, less if you do. FEM has similar requirements, with even more focus on memory performance.

Quote:
Will it be helpful if I use 8*DDR4 3200 memory configuration
For a standard HEDT platform with only 4 memory channels, using 8 DIMMs has no hidden benefit. Maximum memory overclocks on the memory will be lower with 2 DIMMs per channel. But of course you can use more total memory.

Quote:
also want to know the stability of the over clocked memory
The stability of any overclock is up to you. You will have to make sure that the system remains stable with the overclock you apply.

Quote:
And how many CPU cores won't be throttled by such memory configuration?
I would recommend at least 12 cores for Intel. And 16 cores for AMD, TR 2950X is hands down the best value.

Quote:
fluent uses avx2 techonology. So will an intel platform be better?
Fluent may use AVX and higher instructions, but they offer no real benefit here. Memory bandwidth limits how fast Fluent runs, not instructions per cycle.

Long story short: TR 2950X or I9-9920x. Intel has better memory compatibility especially when overclocking, The AMD CPU is slightly cheaper. When comparing the total system cost -factoring in expensive X399 motherboards and compatible memory with overclocking headroom- the price difference is negligible.
mousefire likes this.
flotus1 is offline   Reply With Quote

Old   March 12, 2019, 20:12
Default
  #3
New Member
 
Hao Tang
Join Date: Feb 2019
Posts: 3
Rep Power: 7
mousefire is on a distinguished road
Quote:
Originally Posted by flotus1 View Post
Seems like you need a bit of everything: parallel performance, single-threaded performance and memory bandwidth. So there has to be a trade-off.
For Lightroom (and many other applications): https://www.pugetsystems.com/labs/ar...l-vs-AMD-1298/
The general rule of thumb for CFD: 2-3 cores per memory channel. More if you don't have to pay a per-core licensing scheme, less if you do. FEM has similar requirements, with even more focus on memory performance.


For a standard HEDT platform with only 4 memory channels, using 8 DIMMs has no hidden benefit. Maximum memory overclocks on the memory will be lower with 2 DIMMs per channel. But of course you can use more total memory.


The stability of any overclock is up to you. You will have to make sure that the system remains stable with the overclock you apply.


I would recommend at least 12 cores for Intel. And 16 cores for AMD, TR 2950X is hands down the best value.


Fluent may use AVX and higher instructions, but they offer no real benefit here. Memory bandwidth limits how fast Fluent runs, not instructions per cycle.

Long story short: TR 2950X or I9-9920x. Intel has better memory compatibility especially when overclocking, The AMD CPU is slightly cheaper. When comparing the total system cost -factoring in expensive X399 motherboards and compatible memory with overclocking headroom- the price difference is negligible.
\

Thanks for your reply, problems all solved. I'll compare the Intel(9920X) and AMD(2950x) platform's total price to make the final decision.
mousefire is offline   Reply With Quote

Reply


Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are On
Refbacks are On



All times are GMT -4. The time now is 11:35.