milkyway@home ati/amd app using 0.60 CPU now?

[sangyup81]

This is the reason I don't do PrimeGrid on my GPU. Is there any way to make it go back to using 0.01 CPU like it used to or do I have to switch to Collatz

 

[Sunny129]

i've been crunching MW@H on my HD 5870 for quite some time now, and i've never seen BOINC show CPU usage as low as 0.01...according to BOINC, my MW@H GPU tasks use (and always have used) 0.05 CPU each. either way though, 0.6 CPU is too high for sure...can you confirm via (Windows Task Manager) that an individual MW@H GPU task is actually consuming 0.6 CPU, or is that just what's showing in the BOINC manager? the reason i ask is b/c the value shown in the BOINC manager is really just an average, or an estimate, of how much CPU a GPU task should consume...in reality such a GPU task can consume more CPU (or less CPU) than that estimated value.

also, just as confirmation, my MW@H GPU tasks are running just fine right now...that is to say, my BOINC manager shows each MW@H task as using 0.05 CPUs + 0.50 ATI GPUs (i run 2 tasks at a time). are you running the application as stock standard, or do you have an app_info.xml file in your Milkyway@Home project data directory?

 

[sangyup81]

You're probably right about 0.05. It's not the actual % I'm worried about but the fact that I am now only able to run 3 instances of correlizer instead of the 4 that I used to.

I have no app_info.xml going on and I haven't changed anything since the time it used to use 0.05 CPU. The only thing that changed is that Milkyway went down over the new year and when it came back up, my CPU usage is 0.60 now

 

[GLeeM]

or do you have an app_info.xml file in your Milkyway@Home project data directory?

Is there another kind of file that goes in BOINC directory that can control how much CPU is used for GPU tasks?

 

[Sunny129]

hmm...well if its affecting other projects, then regardless of whether the BOINC manager's assessment of 0.6 CPU per GPU task on your host is accurate or not, it seems those GPU tasks are actually consuming more CPU resources than they used to. do you frequent the Milkyway@Home message boards? you'll probably get more responses if you start a thread over there, and the responses will probably come more frequently too. a quick perusal of the number crunching sub-forum brought my attention to THIS thread in which someone tries to address an issue where his GPU tasks are consuming 50% (0.50) CPU. he tried installing different drivers for his ATI/AMD card and that solved the problem. while i don't know if it'll do the same for you, it might be worth a shot.

also, try this: set MW@H to "no new tasks" and finish all MW@H tasks that are in the work queue. once that's done, detach your host from the MW@H project and then reattach it (be sure to report the completed tasks before you detach though!). see if that doesn't get the CPU consumption of your MW@H GPU tasks back to normal...try this before you try different video card drivers.

 

[Sunny129]

Is there another kind of file that goes in BOINC directory that can control how much CPU is used for GPU tasks?

well just to clarify, the general BOINC data directory is different than the MW@H data directory...i'm sure you're aware of this, but i just wanted to clarify for any others who might not realize that we're talking about two different directories. the app_info.xml i spoke of goes in C:\Documents and Settings\All Users\Application Data\BOINC\projects\milkyway.cs.rpi.edu_milkyway.

as for any files in the general BOINC data directory (C:\Documents and Settings\All Users\Application Data\BOINC) that might limit CPU usage for GPU tasks, i'm not sure. i had to play with the client_state.xml a long time ago, but can't even recall what for...i'm pretty sure it had nothing to do with GPU computing though. the client_state.xml is a file that is installed when you first install BOINC. i've added a cc_config.xml file to my BOINC data directory, but it only uses the <ignore_ati_dev>N</ignore_ati_dev> command line option so i can force BOINC to ignore a specific GPU.

 

[sangyup81]

hmm...well if its affecting other projects, then regardless of whether the BOINC manager's assessment of 0.6 CPU per GPU task on your host is accurate or not, it seems those GPU tasks are actually consuming more CPU resources than they used to. do you frequent the Milkyway@Home message boards? you'll probably get more responses if you start a thread over there, and the responses will probably come more frequently too. a quick perusal of the number crunching sub-forum brought my attention to THIS thread in which someone tries to address an issue where his GPU tasks are consuming 50% (0.50) CPU. he tried installing different drivers for his ATI/AMD card and that solved the problem. while i don't know if it'll do the same for you, it might be worth a shot.

also, try this: set MW@H to "no new tasks" and finish all MW@H tasks that are in the work queue. once that's done, detach your host from the MW@H project and then reattach it (be sure to report the completed tasks before you detach though!). see if that doesn't get the CPU consumption of your MW@H GPU tasks back to normal...try this before you try different video card drivers.

Sunny129, you are a genius! Detached and reattached and I'm back to 0.05 CPU

 

[Sunny129]

glad to help a fellow cruncher

i'm willing to bet that the transition to the new server was the culprit. on 12/27/11, guys who run the project at RPI began transitioning everything to a new server, and estimated that it would be up and down for a week or so before things got steady again. well i guess its taking them a bit longer than expected to iron out all the wrinkles...but when is that ever not the case with DC projects? at first i was having a problem too, but it wasn't the same as yours - my host simply stopped fetching new work out of the blue. i kept getting a message in the BOINC event log to the effect that an HTTP: file could not be found. i found a thread about it on the MW@H forums, and while others were having the same problem, their hosts were eventually finding the HTTP: file and getting new work. i literally had to detach my host from the project and reattach it to get the HTTP: file and get new work. i guess you could say that detaching from and reattaching to a project, or exiting and restarting BOINC are some of the more popular BOINC troubleshooting tricks, just like restarting Windows was always a popular fix-it for various OS-related issues.