Seti task runtime on Westmare or IvB or Hsw/Bdw

Message boards : Number crunching : Seti task runtime on Westmare or IvB or Hsw/Bdw
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile Piotr

Send message
Joined: 24 May 17
Posts: 18
Credit: 20,069,282
RAC: 41
Poland
Message 1882466 - Posted: 6 Aug 2017, 20:11:45 UTC

Hi, curently I'm using double X5675 Westmares. I'm interested how S@H tasks would benefit from AVX or AVX2 instruction sets. Now my 3.07GHz thread takes 4h50mins to process a workunit. How long would it take on E5-26.../V2 or E5-26... V3/4, supposing the same GHz (which I know that are different)? Is it worth transiting to those processors comparing to geting a second double X5675 computer (which is times cheaper)?
ID: 1882466 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1882493 - Posted: 6 Aug 2017, 21:57:22 UTC - in response to Message 1882466.  

Hi, curently I'm using double X5675 Westmares. I'm interested how S@H tasks would benefit from AVX or AVX2 instruction sets.

I can't remember the actual improvement, but I think it was around a 30%+ speedup going to the AVX application on my i7. The speedup from moving to AVX from the 8.08 (Alt) application won't be as great, but it will still be significant.
It is well worth it, however you do need very good cooling to cope with the increased load- I had to replace my heat sink with an after market one as the ambient temperatures here are generally in the mid to high 30°c.
Grant
Darwin NT
ID: 1882493 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1882501 - Posted: 6 Aug 2017, 23:03:15 UTC - in response to Message 1882466.  

Hi, curently I'm using double X5675 Westmares. I'm interested how S@H tasks would benefit from AVX or AVX2 instruction sets. Now my 3.07GHz thread takes 4h50mins to process a workunit. How long would it take on E5-26.../V2 or E5-26... V3/4, supposing the same GHz (which I know that are different)? Is it worth transiting to those processors comparing to geting a second double X5675 computer (which is times cheaper)?

I normally find the AVX app is about 35-40% faster than the SSE3 app on my hosts.
My 16c/32t dual E5-2670 @ 3.0GHz host using the AVX app complete SETI@home tasks in ~2-2.5 hours.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1882501 · Report as offensive
Profile Piotr

Send message
Joined: 24 May 17
Posts: 18
Credit: 20,069,282
RAC: 41
Poland
Message 1882587 - Posted: 7 Aug 2017, 6:50:44 UTC

Good to hear! So according to HAL9k, I can expect to cut processing time nearly by 2. As I see your 32t host is making 30k+ RAC which match with my estimations to my current RAC.
Now I'm thinking a second 2x5675 unit for allmost a peny or go insane IvB/Hsw. Beside considering this, I know that GPU's are the answer :)
In my position heat is not a big deal; I'm on north of Poland and even now in summer my home needs some ~350W of constant heating to keep comfort. So I decided that Seti will receive the benefit of my heating issues (the only way is electricity).
ID: 1882587 · Report as offensive
Profile Piotr

Send message
Joined: 24 May 17
Posts: 18
Credit: 20,069,282
RAC: 41
Poland
Message 1887715 - Posted: 3 Sep 2017, 15:30:49 UTC

Hello, recently I noticed in CPU statistics, 15 Threadripper machines running 4.2 GFlops. So here is my question to X1950 owners, what is the time of standard 183.887 GFlops job on it? Also I read somewhere that AVX instruction executing speed is half on TR's , so are they worth the grand, or better keep with (wait for 8700K) Intel?
ID: 1887715 · Report as offensive

Message boards : Number crunching : Seti task runtime on Westmare or IvB or Hsw/Bdw


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.