I was wondering why it was looking like a task would take 12 days on my 9800 when others have had success. I finally noticed that there was a acemd2 task that showed up at the top of my process chart once every blue moon. I decided to set the niceness of the task to 0 and then the task stayed at the top of the process chart and started making some progress. I finally realized that the task was needing the full core, not just the 0.27 that boinc was claiming it was using. So, I am now baby sitting my boinc client to make sure I only enable one yoyo task at once so that gpugrid can do its thing.

Anyone else running linux have this problem or had it and know of a fix? I do know that the GPU is being used now because the temp goes up. It's at 160F instead of 130F.
____________
Gentoo is not just a penguin.

This is normal for Linux and significantly improves runtime for the GPUGrid task
Let this task complete, to see how it gets on.

Anyone else running linux have this problem or had it and know of a fix? I do know that the GPU is being used now because the temp goes up. It's at 160F instead of 130F.

I'm just crunching an Einstein-Cuda, as well testing my new GT240, and it behaves considerably better.
I know that it's not as good in utilising the GPU, I can see it at the low temperatures of it, but is states quite clear: "Aktiv (1.00 CPUs + 1.00 NVIDIA GPUs)".

So in regard of programming and utilisation Einstein's bad, in regard of behaviour towards other projects Einstein's fine.
Here's the opposite: In regard of programming and utilisation GPUGrid's good, in regard of behaviour towards other projects GPUGrid's bad.

I think it should be very easy to correct the latter one.
____________
Gruesse vom Saenger

For questions about Boinc look in the BOINC-Wiki

Well, I have found a way to keep my yoyo work units to one at a time for most of the time. I just have to estimate the length of time for each work unit and write an 'at' command for each one. There's probably a way I could write a bash script to handle this automatically, but who's got the time?

Perhaps it's possible to find the minimum of CPU % required and fix it with cc_config or app_info on a case by case basis.

I think I have a script that will work for yoyo. Modify for your own secondary project (change "yoyo" to a partial of the one your working on).

Einstein barely uses the GPU so their 1.00 CPUs + 1.00 NVIDIA GPUs is also wrong. On Linux GPUGrid basically allocates one CPU core/thread to facilitate the GPU. This 'bug' may be fixed after Sept but it does speed up the tasks. You might want to consider leaving a core free in Boinc, running GPUGrid, 3 normal CPU tasks, and a light FreeHAL task (they only use about 3 or 4% CPU).

My script above didn't work properly. I fixed it below.

Also, I created an app_info.xml that seemed to work. By claiming a full core, it stayed active and stopped other processes from interrupting. It seemed like it was a little slower, though, so I disabled it.

My script works well enough (almost too well) as long as boinc isn't interrupted.