WARNING: This website is obsolete! Please follow this link to get to the new Albert@Home website!
CPU utilisation factor is very wrong and lets a core sit idle |
Message boards :
Problems and Bug Reports :
CPU utilisation factor is very wrong and lets a core sit idle
Message board moderation
Author | Message |
---|---|
Saenger Send message Joined: 15 Feb 05 Posts: 361 Credit: 600,993 RAC: 0 |
The WUs proclaim to use a whole CPU (1 CPUs + 1 Nvidia GPUs), while in reality it let's it sit idle and thus hurts other project by depriving them off CPU-power. Every time one of those lying WUs crunch, my machine doesn't run at 100% power any more: This doesn't happen with Einstein WUs, that rightly claim only 20% of a CPU. If that proclamation of the use of a whole CPU to my GPU would be even remotely correct, the crunch-time and the CPU-time should be in about the same range, but if you look at my results, they are not that much different from those at Einstein. Albert: Ratio 6700/1000 = 6.7 Einstein: Ratio 6000/700 = 8.6 Can you please change the CPU utilisation factor back to normal, i.e. 0.20, maybe 0.3, but not 1.0? Grüße vom Sänger |
Alex Send message Joined: 1 Mar 05 Posts: 88 Credit: 398,734 RAC: 0 |
There is a workaround available: place a cc_config.xlm into your boinc data directory <cc_config> <options> <ncpus>5</ncpus> </options> </cc_config> Set the ncpus to a value higher than the nr of real cores. |
Saenger Send message Joined: 15 Feb 05 Posts: 361 Credit: 600,993 RAC: 0 |
There is a workaround available: Not for me, as I'm running SimOne as well, a mt-WU, that doesn't like to be run on less cores than said. I lost one because I had that "solution" set. Is there any reason for this weird and destructive setting? Grüße vom Sänger |
[FVG] bax Send message Joined: 19 Jun 08 Posts: 2 Credit: 1,004,500 RAC: 0 |
|
Bernd Machenschalk Volunteer moderator Project administrator Project developer Send message Joined: 15 Oct 04 Posts: 1956 Credit: 6,218,130 RAC: 0 |
Thanks for reporting. We're looking into this. BM |
Alex Send message Joined: 1 Mar 05 Posts: 88 Credit: 398,734 RAC: 0 |
Thanks for reporting. We're looking into this. Hi Bernd, nice to read this. When you (or whoever) is looking into this, please don't forget: Albert currently runs at 50% of possible power. This is how my testsystem runs atm. This is how the cpu-load and the gpu-load is reported. Some weeks ago the settings to run more than one gpu-project at a time were added; unfortunately they work for the nVidias only. Is it really that complicated to change this? If it fails, it's everyone's choice to switch back to 1 wu at a time. I'm pretty shure that it will work; MW runs two wu's at a time, Seti runs two on the APU and three on the HD5830, so what is the reason not to enable this? GPU's are too expensive to run at 68% load. Well, yes, there are workarounds or tricks, but as posted earlier here, they do not work for everyone. I'm shure that will make more than two people happy. Edit: Will the ATI wu's also be moved to Einstein? |
Bernd Machenschalk Volunteer moderator Project administrator Project developer Send message Joined: 15 Oct 04 Posts: 1956 Credit: 6,218,130 RAC: 0 |
Sorry, I'm still ill. Fixing avg_ncpus will take a bit longer. GPU utilization for atiOpenCL plan classes should work now. Will the ATI wu's also be moved to Einstein? I don't understand that question. What are "the ATI wu's"? Can you rephrase it? BM |
Alex Send message Joined: 1 Mar 05 Posts: 88 Credit: 398,734 RAC: 0 |
Shure, I can. Will we see wu's for ATI-cards @ Einstein sometimes? THX !!!! Sorry, I'm still ill. Fixing avg_ncpus will take a bit longer. No Problem, health first. Gute Besserung und Frohe Ostern! Edit: GPU-load is 88% now, much better than before! |
Bernd Machenschalk Volunteer moderator Project administrator Project developer Send message Joined: 15 Oct 04 Posts: 1956 Credit: 6,218,130 RAC: 0 |
Will we see wu's for ATI-cards @ Einstein sometimes? Albert is not a production project on its own, it's just a project for testing things that we intend to run on Einstein at some point, but can't test there without putting our production project at stake. So in principle everything that we are doing on Albert sooner or later is meant to end up on Einstein. The BRP workunits on Albert don't differ much from the ones over at Einstein. What differs are the applications (or in BOINC terminology: application versions), and some server components (scheduler, validator, possibly workunit generator, database). This is what's currently being tested on Albert. As this very thread shows, there are still issues with this server code that need to be solved before we could move this over to Einstein. BM |
Bernd Machenschalk Volunteer moderator Project administrator Project developer Send message Joined: 15 Oct 04 Posts: 1956 Credit: 6,218,130 RAC: 0 |
Should (finally!) be solved. BM |
ihatelolcats Send message Joined: 3 Jul 12 Posts: 4 Credit: 9,030,454 RAC: 0 |
WUs I have are showing 0.898 CPU. Can this be a user configurable option? Or make it something like 0.75. Thanks |