I've just noticed that the two units that I have completed on my
660Ti have the same amount of CPU time as GPU time.

Is this correct?
I also notice that despite Boinc showing (0.03 CPU and 1 Nvidia GPU) that according to my resource monitor, ACEMDLONG is averaging 12.5% cpu, which, on my system, is a full core (i7-2600K)

http://www.gpugrid.net/result.php?resultid=6111707

http://www.gpugrid.net/result.php?resultid=6043519

Are these tasks just using a lot more CPU than they used to, or do I need to change something??

Also, I am running 8 Docking tasks at the same time, I never worried about leaving a core free before because GPUGrid used to use so little CPU and leaving a core free did not make any difference to the runtime.

I thought this was normal... in my case BOINC shows 0.615 CPU + 1 NVIDIA GPU. It basically takes a whole thread (HT'ed CPU). I let BOINC use only 88% of my cores as to leave one thread available for GPUGRID, doing this, the task manager shows 100% CPU usage.

Also, freeing TWO threads increases the GPU utilization from 96-97% to a solid 99%, but I rather crunch another WU with that thread at the expense of 1-2% of GPU usage.

CPU: i7 3610QM
GPU: GTX 660M

I'd imagine the faster the GPU, the more CPU time it'd need to keep it fed... so yours should use even more CPU than mine.

True, but the never used to be equal.
it might have been 20 to 1 in favor of the GPU.

no big deal.

GF600 uses up to 1 CPU core/thread.
GF500 and earlier don't.

The respawn tasks seem to use about 75% of a thread for me, while the NATHAN_RPS tasks use 1 full thread. An app_info stating to use 2 threads for these might expedite these tasks.

If you have a GTX600, free up one CPU core/thread. This is recommended for 4core/8thread processors anyway.

Why? There are probably some processes that now perform faster on the CPU when using a 600 series GPU. The shaders are more pleantyful but also a lot leaner.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

I tried freeing up a thread, but it didn't make any difference to the runtime that I could see, nor did it change the GPU usage (96%)
Probably because of the lack of GPU scheduling and that I run my CPU at 4.5 Ghz.

It is interesting that these tasks on a Kepler card use so much more CPU time than on a Fermi.

Use lots of Vram too, They seem to take around 1GB.