Originally Posted by Krusher33
I've been all "to hell with it" and doing 4 units.
Same here. I used an app_info.xml file
to run 4 tasks on my cards.
Originally Posted by Doc_Gonzo
I've got 3 tasks per card going on all 4 cards. 2 tasks = 28 - 30 minutes. 3 tasks = 36 - 38 minutes on the 7950's. I'm giving them 0.25 CPU so I get 7 x WCG work units going on the 3770K and 3 x WCG work units on the 3570K. It's weird - the WCG work units complete in around 3.5 hrs on the 3570K and take around double the time on the 3770K. Maybe because of the hyperthreading being weaker than a full core? I've only had the i7 for a few days so have no experience with this
HT makes the cores "weaker" by splitting some of the internal hardware to work with two things at once, but as you do more work, it's usually left on as it yields better PPD. None of my computers has HT off.
My GPU tasks take roughly ~50 min to complete on both my 7950 and 7970, using 0.25 CPU per WU, so 4 tasks take only 1 CPU core/thread (you can modify that on the app_info file)
Originally Posted by Wheezo
Two seems to be ideal for my 7870, runs at 100% as long as it gets enough cpu.
You might wanna try 3. My GTX 470 is happily crunching that and maxing the VRAM. Maybe the limit for Einstein is the VRAM, not too much the cores, but yeah, there should be a limit on how much you gain when running multiple WU's per card.