I know BOINC allows CPU throttling by percentage.
Is a like function workable/possible on CUDA devices?
It is something I would love to see.

What result you would like to achieve, you question is too common. It would be better explaine in details, for example.

What result you would like to achieve, you question is too common. It would be better explaine in details, for example.

Recently I joined gpugrid and find it to "overwhelm" my system. Meaning, my system becomes unresponsive for about 45 seconds every 2-4 minutes.
Perhaps having throttle in place would allow client to reduce/limit of GPU use.

I totally agree on the idea of having a way of throttling the GPU.

Although if there is a way to make sure that you are only using 50% of the GPU... that I do not know about... then I am sorry.

This is an incredible site and program, just thought I'd give my two cents worth.

You could underclock the GPU, but that will not free up the system.

Try reducing the CPU usage - System responsiveness will be more to do with that.

You could set Boinc to only use the CPU 80% of the time, or perhaps just use one, two or three cores - for those that have a multi core CPU.

There is a fair chance that other non-Boinc and non-GPUGrid things are responsible for locking up your system:
Updates, web pages left open, Outlook, Backups running, Playing online content or watching a film, Disk defrag or scan by AV software!!!

IMO there is no need for throttling the GPU: Try to look at your card with gpu-z. On more recent cards like GTX260 and up you'll see that GPUGRID only use up to 60% of the GPU anyway. Collatz will use around 85% constantly while SETI has an average of 55% with spikes up to and higher than 90%.

IMO there is no need for throttling the GPU: Try to look at your card with gpu-z. On more recent cards like GTX260 and up you'll see that GPUGRID only use up to 60% of the GPU anyway. Collatz will use around 85% constantly while SETI has an average of 55% with spikes up to and higher than 90%.

Just noticed the same with the new version of GPU-Z. I'm used to having a 90%+ load on my ATI-cards, and wondering if there are any settings I could try with GPUGrid applications? Seems like a waste to just use 60% of available power. With the ati apps you can define parameters or run two GPU-apps at the same time to maximise efficieny.

Well, if you have something like a 9800GT or, even worse a Quadro NVS290 , the load on the GPU will be much higher. That means that the GPUGRID - WUs doesn't scale with better hardware. The gain in speed is partly through more shaders and/or higher speed and/or the wider memory interface.

And all one can do about it is build up pressure on the project team to get the programming right or do it yourself (Note: I even don't know if the source is somewhere around).

BTW, a software, build into the driver of the card, which would allow to assign time slices to applications running on that card wouldn't do any good. As the applications itself use only part of the capability of the hardware, you only get more overhead. On the other hand, a thing like Hyperthreading would do the job.

But I doubt seeing something like this from Nvidia.

I found that TThrottle was designed just for this problem, although it did not work on my 9800XT... it is doing a amazing job on my 9800M GS.

The GPU is running right at 158F/70C, which after digging into it, that seems fine.

Here's the URL:

http://www.efmer.eu/boinc/

Hope this helps