Message boards :
Number crunching :
Too big runtime differences using 2 PCIe 2x ATI 5870 GPUs and an i7-2600
Message board moderation
Author | Message |
---|---|
Fred J. Verster Send message Joined: 21 Apr 04 Posts: 3252 Credit: 31,903,643 RAC: 0 |
For sometime I noticed, device 0 is faster, having a (much) higher GPU-load, the 2nd device 1 hadn't even half the load, ~85% v.s. 4.1%, thus higher runtimes. And I remembered running MW work, freeing 1 core, 2 threads (HT=on), loading the GPU took 3 seconds. Which started cooling problems, cause no time, to cool of, the higher loading speed and extra thread gave a 100% load. With no 'free-cores', loading took 13 seconds, tried it on SETI and it looks like the same 'problem'. So freed 1 core, 2 threads and load in 2 GPU are almost identical. CPU is an i7-2600, GPUs are ATI (AMD) EAH5870s, PCIe 2.0 x8 slots, PSU=1KWatt ATI Host; i7-2600 CPU. I could try core-affinity, but this works, also changed to 2 instances _per_device. (Also disabled smartdoctor, which started after BOINC was running). (Also use a 32GByte CF card as ReadyBoostCache, no HDD access, running only BONC {7.0.25}). Could install BOINC on the CF card elimenating HDD access, but BOINC 'tells' windoze what to do when and how, which will trigger commands and useally HDD access. Don't know if it makes a (noticeble) difference? |
Mike Send message Joined: 17 Feb 01 Posts: 34257 Credit: 79,922,639 RAC: 80 |
Thats a known issue with recent ATI drivers Fred. With each crime and every kindness we birth our future. |
Fred J. Verster Send message Joined: 21 Apr 04 Posts: 3252 Credit: 31,903,643 RAC: 0 |
|
.clair. Send message Joined: 4 Nov 04 Posts: 1300 Credit: 55,390,408 RAC: 69 |
I think i am seeing a similar slowdown if i load all core`s, using ccc12.4, The card came with ccc11.2 on its CD, and that one used lots of cpu and not much gpu. I am having to keep one core per WU free to feed my 7970 with two WU running That gets me about 90% gpu useage, if i try to run three WU on the gpu the runtimes get very long, and gpu useage does still not reach 100%. Rosetta gets to play with two cpu cores and Freehal nci get`s to butt in on the gpu feeder cores, it seems to work ok, |
Mike Send message Joined: 17 Feb 01 Posts: 34257 Credit: 79,922,639 RAC: 80 |
I´m running 3 instances on my HD 5850 and keep 2 cores free since the FX uses shared FPU. I always have 99% GPU load. With each crime and every kindness we birth our future. |
©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.