Is there anyway to automatically set the affinity for WU's so they don't keep going back and forth between the cores. There is some gains (~5%) to be had by doing this, I know Crunch3r's optimized Boinc client did this however I cannot get it to run (always error when try to restart Boinc service). Maybe the Boinc team could have a look into this for an updated Boinc client.

I agree with you that using affinity is better. I have experienced Crunch3r's affinity optimisation at work...so I don't understand why the Boinc team can't get it to work? It would be an even more benefit now with cuda GPU projects.

---

MAIN PC:
Q6600 @ 2.40GHz - Windows 7-64Bit 4GB RAM - BOINC 6.6.38 Win64 - 9800GX2 190.38
NETBOOK:
N280 @ 1.66GHz - Windows 7-32Bit 1GB RAM - BOINC 6.6.38 Win32

I was wondering if there is a reason why the BOINC Team have not put CPU/GPU affinity into BOINC?

Crunch3r could do it in in his last optimised client, so why can't the Bonic Team.

---

MAIN PC:
Q6600 @ 2.40GHz - Windows 7-64Bit 4GB RAM - BOINC 6.6.38 Win64 - 9800GX2 190.38
NETBOOK:
N280 @ 1.66GHz - Windows 7-32Bit 1GB RAM - BOINC 6.6.38 Win32

...so I don't understand why the Boinc team can't get it to work? It would be an even more benefit now with cuda GPU projects.

They never included it and as far as I know they never will as the benefit of 5% is not worth it.

There has been a whole discussion about the pros and cons of CPU affinity in the Einstein forums. The people that tested it under controlled circumstances, could find no real benefit over BOINC without CPU affinity. Those that do use it in Crunch3r's client, can't or won't give real evidence when asked although they'll keep saying it's better for them, no matter how much testing by others refute their claims. (Besides, it is Windows only. If something like this was to be included it would need to work on ALL platforms without too much code-tinkering.)

As for CUDA projects, there's not that many around. 2 have it in testing. The client that can do CUDA isn't stable yet, it even falls back on using a CPU in combination with the GPU. So the benefit claim can be refuted immediately.

Come back with the request in another year when there are more projects doing CUDA, when ATI is also supported and the clients that have been out there have been stable enough.

---

Okay benefits for me with affinity; if I have 3 WU running on my quad core with affinity, when one WU finishes, the next WU would normally start on the 4th core. I guess I prefer my cores to have rest/cool down breaks.

Im running both GPUGrid & the "Seti Beta Cuda". Currently I am running 3 CPU WU & 1 GPU/Cuda WU, and as the latter never uses the whole CPU it would be nice if each WU had a dedicated core to run on.

In total I have 6 projects, and I always thought that it was better the each core crunched only 1 WU each, not all 4 cores crunching parts of each of the 4 WU running.

Also it would make more sense that if a user only want 3 of their cores to be used by Boinc, that only 3 cores are used, because at present, it will still use 4 cores (75% of each core), even if you choose 3 cores.

---

MAIN PC:
Q6600 @ 2.40GHz - Windows 7-64Bit 4GB RAM - BOINC 6.6.38 Win64 - 9800GX2 190.38
NETBOOK:
N280 @ 1.66GHz - Windows 7-32Bit 1GB RAM - BOINC 6.6.38 Win32