12 simoultanous tasks running at 99% on 4870x2
Message boards :
Number crunching :
12 simoultanous tasks running at 99% on 4870x2
Message board moderation
| Author | Message |
|---|---|
|
Send message Joined: 31 Mar 08 Posts: 61 Credit: 18,325,284 RAC: 0  
|
<avg_ncpus>0.1</avg_ncpus> <max_ncpus>4</max_ncpus> <cmdline>n8</cmdline> http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=26311777 I'm trying to push the card to a crashing point, considering the recommended setting is 256MB per wu, but too bad the server is dry and so far no crashes! I've noticed GPU time is 30 - 40 seconds weather it was one task or multiple tasks! |
 Send message Joined: 8 Nov 08 Posts: 178 Credit: 6,140,854 RAC: 0 
|
<avg_ncpus>0.1</avg_ncpus> No, it's lower than 256MB/WU. I've inadvertently had 30 tasks running at once on my HD4870 1GB, which works out to less than 33MB/WU at least. This was with the 0.17 or 0.19b application, no 0.19d though, so I don't know if there's any change with that. 
|
|
Send message Joined: 26 Jul 08 Posts: 627 Credit: 94,940,203 RAC: 0  
|
No, it's lower than 256MB/WU. I've inadvertently had 30 tasks running at once on my HD4870 1GB, which works out to less than 33MB/WU at least. This was with the 0.17 or 0.19b application, no 0.19d though, so I don't know if there's any change with that. It depends on the WU. The ~18 credits WUs should need about 28MB, the 27-29 credits WUs about 37MB on the graphics card irrespective of the application version. But one should not load the card to the max as this leads to slowdowns (reduced cache hit rate, swapping over PCI-Express) or even errors (if there is simply no memory left) and does not help the performance. Two or three (the default value) concurrent WUs are really the best to get maximum throughput on your GPU. |
|
Send message Joined: 31 Mar 08 Posts: 61 Credit: 18,325,284 RAC: 0
|
I'll stick with 3 then! If I'm to believe catlyst, 1 gets a below 50% utilization, 2 a 70%, and 3 fluctuate between 80-99! |
©2024 Astroinformatics Group