Just now I tried to crunch for GPUGRID on GTX275 (linux x64), using 190.18 driver and CUDA 2.3. The problem is that GPUGRID really using CPU time a lot (jumps to 100% every couple of seconds), so Rosetta started to calculate each WU for 5-6 hours instead of 3 hours. I left GPUGRID, but I want to contribute my videocard to the project. BOINC client version is 6.6.36 and I've got it from here.

What should I do?

stupid question, pls ignore and delete this topic.

as far as I understand GPUGRID for some reason needs to use CPU. My intension was to run Rosetta at full speed and left the rest to GPUGRID. But while Rosetta uses 100% of CPU time, nothing left for GPUGRID.

So, I can live with it :-)

I don't want to discourage you from calculating for Rosetta but your video card will do many more times useful work for Gpugrid than your CPU can do for Rosetta.
A good solution would be to use Gnome System Monitor to change the priority of either Gpugrid or Rosetta. You can assign more priority to whichever project you like. The other one will still get CPU time, just less.
On my system, I set the Gpugrid application to maximum priority and Aqua to slightly lower priority than Gpugrid. The Gpugrid cuda application is much more efficient than a CPU floating point calculation so I want it to have higher priority. You will have to reset the priorities as each new work unit comes on line.
Hope this helps.
What version of Linux are you using? I am currently using Ubuntu 9.04 but I have also used Sabayon and many versions of Suse.

I'm using Ubunutu 9.04

Priority. GPUGRID running an nice 10 while Rosetta at nice 19, so it OK :-)

Another question to u. On top of GTX275 I've got 9800GT which I'd like to contribute to the project also. The only problem is that my mobo (Asus P5Q-PRO) do not support SLI (as least now, but rumours about SLI patch for P45 chipset are floating).

So, Can I stupidly insert 9800GT into mobo or there some trick - in terms of driver (I'm using 190.18), GPUGRID or whatever else?

I don't want to discourage you from calculating for Rosetta but your video card will do many more times useful work for Gpugrid than your CPU can do for Rosetta.

Hi Betting Slip, it depends on how you read his comment. I took it as an explanation why it makes sense to use more priority on the CPU for GPUgrid to feed the GPU as to maximize the benefit for both.

Chevy and Ferrari drive on the same road, it makes sense for the chevy to drive on the right lane, so that the Ferrari can overtake it safely on the left (mainland driving rule) and be used to its full potential. Neither is hindered. It does not judge the necessity of the Ferrari to drive fast, nor the reason why either of them are on the road or where they are going.
Just a commonsense decision.

OK, guys, calm down pls :-)

I'm not desputiong about the priorities, I thought that there is an easy way to split resources of my computer between the projects in order to donate it on the most effective and smart manner. Nothing more but nothing less :-)

I just came from F@H and there GPU client using 2-3% max of CPU time, so I though I can do the same trick here. It's not the case AFAIK, but it's not big deal.

I've got 2 questions. I'm running ubuntu 9.04, nvidia GTX275 OC'd by me (702/1584/1260 - absolutely stable), desktop effects (compiz, avant) on, nvidia driver 190.18 beta with CUDA 2.3 & BOINC 6.6.36.

1. From time to time my computer stucks but after some time (from couple of seconds to 20-30 seconds) it comes to normal. I suspended GPUGRID WU's and this never happens to me again, so the reason is there. I unchecked "Use GPU during work", I also changed the time period (up to 10 minutes), but nothing helps: GPU continues to work, GPU time runs and progress bar moves ahead.

So, how I can actually stop GPU WU's during that time I'm working on PC? Click "Suspend" all the time is a "bit" boring :-)

2. Does BOINC and why need to be run under super user only? What I can I do in order to stop this? The problem is that due to necessity to input super user password in comsole I can not make an icon on avant to start it easily and w/o terminal. If I'm starting it as user it can not connect to localhost.

Thanks in advance and hope it's not too many questions :-)

Perhaps it would be more accurate for GPUGRID tasks to require 0.2 or 0.25 CPUs (with 1 CUDA processor), instead of the current 0.03 CPUs. I know you can set this value for the tasks.

There were some initial release issues with Fedora 11 (what I'm running), and there are two outstanding issues:
1) The /dev/nvidia0 device loses its permissions upon reboot, causing CUDA device not to be recognized;
2) The graphics don't work for any application (probably due to #1);

But back to the original post concern....
The load average of my system running 1 CPDN and 1 CUDA task is 1.45. This is telling me that the CUDA task is using 45% of 1 CPU and (of course) 100% of the video card. Why does my lil' ol' 8600GT need so much CPU support?

My suggestion on favouring Gpugrid over Rosetta is not because one project is better than another or produces more credits. What is undeniable is that mathematical calculations done with a Gpu are at least 8 times faster than those done with a CPU due to the construction of the two different processors. This is true whether the project is Folding@home, Gpugrid, Seti or Einstein.
This is why I favour supporting projects that use GPU processing - more processing power.
Rosetta is an excellent project. So is World Community Grid. Find the one that you like and crunch away.

There is a way around having to run BOINC and Gpugrid as a superuser.

See:

http://einstein.phys.uwm.edu/forum_thread.php?id=1913

It has a discussion of how to edit the gui_auth_rpc.txt file to get the result that you want.

DJStarfox
In fact in my case (Pentium E6300 2800 stoc OC'ed to 4000) it uses 28%-35% aven if it says about 18%...

ExtraTerrestrial Apes
"Do not..." looks do not working either on linux either on 6.6.36, later this week I'll try to install 6.10. But how I can bring all WU's from old client to the new one?

regarding the starting the client w/o SU rights. looks strange coz not any single my program (not system ones) on my computer requires this. I need to spend some reading the forums...

DJStarfox
I've got GTX275 and it uses less then yours 45%, but still smth. As far as I understand nothing much we can do right now...

RalphEllis
I read somewhere about the way how nvidia did double precision calculations on GTX200 family cards. If u'll start to to calculate with it your card becomes very slow. But for CPU it's not an issue, in fact F@H and rosetta using double precision calculations. It gins way more to the science rather than single precision...

But I'm donate my video card to those people who can use it for science, like GPUGRID :-)

http://einstein.phys.uwm.edu/forum_thread.php?id=1913

It has a discussion of how to edit the gui_auth_rpc.txt file to get the result that you want.

I've got 6.10.4 version of the client. works fine for me.

In the same time I fixed the problem with starting the client under SU. Just get to the place where BOINC located and type:
sudo chown -R [username] BOINC

As a bonus I've got the chance to change nice of GPUGRID process in gnome system monitor. Right now I'm testing how it works. Freezes are still there, but more rare and for short time, not for for 20-30 secs as it was before. I made nice=18, a step higher then rosetta but still low enough to avoid freezing the system.