Hello,
I get a gpu usage of only 60-65% on this task:
http://www.gpugrid.net/workunit.php?wuid=1966254

I also seems to take longer to complete it than other tasks.

60 to 65% is probably about right for that task type under Win7.

Some weeks ago I did some testing with Einstein's cuda-app. This app uses the gpu only between 9 and 15% and I have the feeling, that its a waste of resources to run an app like this.
I modified the client_state.xml (section for Einstein), setting the <count> there to 0.22
This had the effect, that I could run up to four einstein wu's at a time on my quadcore machine. GPU-usage was the around 60%. Not a single wu did not validate.
Modifying the client_state.xml is a temporary method. Any connection to the project resets this setting, so using an app_info.xml is a better way to do so.
I tested also: 2 Einstein + 1 Seti cuda wu's, worked OK
1 Einstein and 1 Milkyway, worked also fine. 2 Einstein and 1 MW caused the MW wu to time out.
I'm not familiar with the gpugrid app since my GTX260-192 fails to run this. But looking back to my tests, more than one gpu-app at a time should be possible. Make shure, that the total <count> of all active gpu-apps is <= 1.
For testing I found it good to download a bunch of wu's, disable the connection to the project (no more wu's), stop boinc, modify client_state.xml, restart boinc and be happy about a better gpu-usage.
For permanent use you have to write (or use) app_info.xml's.

Alexander

Interesting post Alexander.

On a Fermi I would expect it might work with a project here and one elsewhere, possibly even two projects here at the same time, but I have not tested this. I doubt this would work here for other cards though. I would expect GPUGrid tasks to fail; although I managed to run a GPUGrid task on a GTX470 at the same time as running a Folding@home task (non-boinc), on other cards it failed the GPUGrid task. The performance for GPUGrid also deteriorated substantially, so I don’t do this.

With XP and Linux the GPU utilization is quite high for most tasks, but you might be able to eke out some additional performance. Of course it could be counterproductive too.
With Vista and W7 the performance is typically between 10 and 15% reduced for all tasks even after optimization, and trying this might not change much.

My issue with Einstein is the research rather than the poor performance - it's furter down my list that GPUGrid.

hey everyone,

I tried the client_state thing, and I was able to fit 1 gpugrid and 2 einstein tasks on my gpu without problems (makes it more productive!)

I managed to install a app_info file on einstein, can anyone please help me create one for gpugrid too?

thanks

EDIT: I still have some unused gpu ressources on that card, so gpugrid won't be slowed down, meaning I'm really running an optimal configuration

You need to check your memory usage as well. GPU usage is not the only limit.
I usually check that with MSI Afterburner.
Alexander

i have done that, this limited me to 2 einsteins, otherwise i could have run 4-5 einsteins