Ok guys, I'm relatively new to CUDA and have a couple of newb questions ;)

I currently have a quad core crunching with 3 CUDA cards. Therefore 3 cores are crunching AP with 3 GPUs crunching MB.

The 1 core dedicated to the GPUs seems to be insufficient to distribute the load to all 3 as GPU crunching speed has fallen noticeably, I'd say about 50-75% slower per GPU.

Is it possible for me to setup BOINC so that only 2 CPU cores crunch AP, thereby leaving two to handle any load for the GPUs? I tried changing the cc_config.xml to <ncpus>2</ncpus> but that seems to have had no effect.

For that matter...is it worth crunching AP at all if it slows the GPUs that much.



Thanks guys.

never mind... ;)

Just set the "On multiprocessor systems, use as most xxxx% of the processors flag to 75% and that did it "rolleyes". Knew it was something obvious.

never mind... ;)

Just set the "On multiprocessor systems, use as most xxxx% of the processors flag to 75% and that did it "rolleyes". Knew it was something obvious.

Your use of "cc_config.xml to <ncpus>2</ncpus>" should have worked, too. If you do want to try it out live, you can edit your cc_config.xml then go to BOINC Manager, Advanced, then select "Read config file". Then hands off of the keyboard and mouse for a minute while it does a re-benchmark.

Thanks - I had to roll my eyes too. I never thought of trying the "use at most xxxx%" approach. Too obvious, I guess. ;)

bob

---

Opinion stated as fact? Who, me?