I wouldn't call myself a F@H Guru but I have bounced between a lot of CPU + GPU combinations folding...
Quote:
Originally Posted by MARSTG  what is the best combination of SMP anf GPU folding in the same time? |
I think even with the lesser bigadv penalty, the best "combination" for points / power consumption in dedicated folding, would actually be a Core i7 2600k OC'd like crazy on a Z68 board with the IGP giving you the desktop. Unless something changed, a 2600k on its own can outproduce any GPU for the same wattage. Now if you HAVE the GPU and CPU already, you might as well fold on both.
Quote:
Originally Posted by MARSTG If you do SMP better leave a core of your multicore CPU free for the GPU client too? Like let's say you have a 6 core Phenom II, you SMP fold with 4 cores and leave 2 other cores for managing the GPU clients, or how is best (yielding best production together)? |
Only if you GPU is folding ATI GPU2 or is a low-end nVidia GPU.
If you have a low-end GPU or are using the older ATI client, the GPU folding will eat a sizeable portion of one core's CPU usage. If you see more than ~5% CPU usage out of your GPU clients combined, then leave a core free for these clients. From what I've seen, if >10% of a core is being used up, the SMP points generation drops so much that you might as well not fold on that core. Recall a WU's behaviour varies a lot depending on the platform you are using + the WUs themselves are different, so you may have to experiment.
What you can do is complete a WU with SMP only, and then try again with SMP + GPU client(s). See if the GPU folding can make up for the penalty you take from the SMP getting disrupted. If you're folding mid to high-end nVidia cards, even from older generations, you should easily produce more GPU points than you are penalized SMP points.