cancel
Showing results for 
Search instead for 
Did you mean: 

PC Processors

Gabo214
Journeyman III

Ryzen 3990x multiple process issue

I have a program that I need to run many times (thousands) to get a Jacobian, I usually do this on Intel processors, without major problems (negligible performance loss), so I expected to have a similar behavior on this new processor, but it doesn´t. I have a significant performance loss when adding a new process, as an example, 1 process takes 30 minutes (approx), 2 processes take 40 minutes each and thus getting worse and worse, reaching run times of more than 6 hours at a time when running 40 processes simultaneously.
I have tested under windows 10 and windows server 2019 (linux pending), I 've recompiled the code under some optimization for AMD processors, and i have made tests setting affinity to determinated cores (to avoid cores switching while running) but nothing seems to work. On the other hand, I've done the same work on intel processor (i5, i7, Xeon) where there is not a noticeable lack of performance while reaching the same amount of process as cores. 

These tests were carried out on 2 different computers, both on 128 Gb ram, NVMe PCIe SSD, dual quadro RTX8000, etc...

Does anyone have any advice or recommendations? I find hard to believe that a 64 cores processor had a worse performance (for this particular case) than a 12 or 18 cores intel processor. 

0 Likes
3 Replies
CBO123456
Journeyman III

Hi Gabo, 

We are experiencing similar issue on our side.

We tried to launch multiple instance of a single program to be run in parallel and we are experiencing significant performance loss when adding a new process. 

 

We tried to set ON/OFF the SMT but it doesn't seem to help.

Did you managed to fix this issue ? If yes would you mind to share teh way forward ?

 

Best regards,

 

 

 

0 Likes

I'm just a newbie when it concerns writing code, but I think the optimization for Intel is written into your program and that is likely jamming up the AMD.

Don't they use different architecture and processing for parallel programming?  Making command structure and the like different in how the code is written to handle processing and CPU time.

I know with certain intel specific commands they can be crunched and run through an AMD processor, but it has to work its butt off to get it done.  Same is true for AMD specific command through an intel.  

Aren't the parallel processing architectures different enough they have to be optimized for the processor?  Just a question from a new guy so to speak.  I would really like to know because I'm betting that is it from the outside looking in.

 

0 Likes

Look at archived thread 

AMD Software Optimization Guide for Ryzen

specifically, a post by reneg 06-28-2018 11:19 AM

It outlines how the same process compares on a 2700x to an intel Skylake.

He posted on the last code block, 3GHz Ryzen 7 2700 is now 1.5x faster than Skylake 3.4GHz

Look at the 3 iterations and how it is laid out for AMD to make it faster.  May help.  Maybe.

Maybe that will make sense.  

0 Likes