If I have BOINC running the system slows down.
It is an office computer with a GeForce 8600 GTS 256MB, a Intel E7300 and 2GB RAM and Windows XP SP3.
The CPU load is always below 5% and more than one GB RAM is free (according to the Task-Manager). Is the system slow because of using the GPU at 100%? I don't know how to see the GPU load.
I would like to use the GPU all the time and the CPU after 3 minutes inactivity.
Is this possible?
If the problem is the 100% load of the GPU is it possible to use only 80% from the GPU while I am working at the computer and 100% on inactivity?
Thanks in advance.

Unfortunately to use the GPU the CPU still gets a small amount of use. Don't think you can get around that at all.
As you have one of the low end cards in that computer the one problem i can see is the card having trouble rendering your screen while crunching away. That could be your slowdown problem. I'm surprised that card makes deadlines. Could
you possibly set the app to only work when the computer is not in use and still make deadlines? I'm not sure myself my slowest CUDA capable card is a GTX280 all my slower cards cant do CUDA anyway.

Edit: I can see that a suspend GPU work while computer is in use has been added to options but wont be usable until BOINC 6.7.x and even the dev versions are 6.6.x so that wont really help you yet. Possibly you could do some Seti GPU work (make sure you adjust your app choices to only do enhanced on CUDA) with this machine and suspend all work when computer is in use if you cant make deadlines here doing the same.

OK, thanks.
Now I have set it up to run only on inactivity but the CPU is not used many times.
After two hours only 1 minute CPU usage. Is it not possible to use CPU and GPU together?

Is it not possible to use CPU and GPU together?

The GPUGRID applications only run on the GPU with brief polling of the CPU. There is no CPU application at this project. If you wish to run both, you will need to add another project that has CPU applications (this is pretty much any other project).

OK, thanks.
Now I have set it up to run only on inactivity but the CPU is not used many times.
After two hours only 1 minute CPU usage. Is it not possible to use CPU and GPU together?

Your computer is slow because of using GPU, I assure you. I ran the projects like einstein@home, folding@home etc. at very slow processors (much slower than yours) and there was no difference in performance.
There are few options. You can suspend project manually when doing something or you can get used to it. In future boinc versions there will probably be an option that uses gpu only when idle.
You can also use some project that only uses CPUs (most projects do) and you can safely put them on work all the time, not just idle time, you will not see the difference.
Of course, you can also get a better GPU. :)

I would like to comment that some projects are not running smoothly together with gpu(ps3)grid.
Those which i know are seti@home and CPDN "Climate Prediction" these can make the project error out.
Seti because it has also a cuda program and CPDN because of very memory hungry units "these can be turned of under settings in your account at their site"
But i would recommend a simple cpu project, many which work very fine with this project.

I would like to comment that some projects are not running smoothly together with gpu(ps3)grid.
Those which i know are seti@home and CPDN "Climate Prediction" these can make the project error out.
Seti because it has also a cuda program and CPDN because of very memory hungry units "these can be turned of under settings in your account at their site"
But i would recommend a simple cpu project, many which work very fine with this project.

Shh! don't tell my machines that.

They seem happy to run Seti (cpu and cuda) at the same time Einstein (cpu only) and GPUgrid (cuda only) work. The cuda jobs tend to run in deadline order, so gpugrid gets in first because of the 4 day deadlines, but its happy to put them into "waiting to run" status when it needs to. Using BOINC 6.6.17 which certainly helps. I can't comment on CPDN as I don't run it.

---

BOINC blog

Well, in case of CPDN he ran several WUs which each required 1.5 GB of RAM. This can cause serious issues even without GPU-Grid, so I'd say it's not a matter of "GPU-Grid & CPDN", but rather an issue of CPDN itself.

MrS

---

Scanning for our furry friends since Jan 2002

That might happen if you run a paticular model at CPDN - it needs 1.5Gb (the HADAM3 Models). The others are fine, even HADAM3's are usually ok if you have 4Gb RAM, and there are also selection boxes in the Project preference pages to state whaich type of model you want. The HADAM3 selection box has a red typed warning re RAM requirements.

I run both SETI CUDA and CPDN, and well as fitting in a GPUGrid model every couple of days, have never had any Project clash.

Regards
Zy

... have never had any Project clash.

Yes, that's what I wanted to say: It's not a project clash. If you tell CPDN to use more RAM than you actually have, you create a problem even without GPU-Grid.

(And it was good you pointed out that CPDN handles these WUs + requirements well, i.e. the selection and the warning).

MrS

---

Scanning for our furry friends since Jan 2002

Guys i can tell you these CPDN units use minimal 1.5 Gb ram up to even 3Gb ram at certain moments, now if i had only 1 cpu it would not be a problem at all.
The problem starts when you have like me multi cores running CPDN, then 4 times this huge amount is being processed and there we have the problem.
Its very easy to control this by babysit cpdn to run only on max 2 cores whilst running other projects.
And i guess a X64 version Os would help also, so when all 60 days models are done on my system ill boot into a X64 OS and use my 8Gb system memory ;)
Second i am gonna change the size of swap space on my system to 12 Gb making it have a bit more available if CPDN is hungry again.
My problem was simply that 3 high memory units where running with a 60 day unit together with PS3grid while it just had limited memory available.
Its simply in win32 you have only 3.2 Gb from 8 available in my system, you need to count the additional memory for the os itself and your programs and services also need memory.
All this combined made my system run out of memory for processing so made gpugrid to crash.