I just got a new machine with a much better GPU than I need with the explicit purpose of doing GPU work, which I hadn't been able to do before. I'm not necessarily looking to take my machine to the limit or spend a ton of time optimizing, just looking to make good use of the resource.
Are there any rules of thumb for preferences/configs for a casual cruncher?
Similarly, are there any rules of thumb for what GPU stats/combo of stats translate into a maximum responsible GFN project? (e.g. a pixel fillrate of X means you probably shouldn't try WUs for anything above GFN-YY)
I really am interested in the rules of thumb, but here's more about my specific case if you're curious:
I found that GFN-18 WUs move excruciatingly slowly, but GFN-17-Low cruises along just fine (we're talking an estimated time 500x greater after letting it run for a bit). That's a little confusing as my admittedly basic understanding (from here https://www.primegrid.com/forum_thread.php?id=6491) is that both use similar calculation methods, and the recent average times suggest a difference of 2x-4x longer. It's probable that I'm not understanding how the CPU settings affect GPU tasks: I'm maxing 2 tasks (1 GPU & 1 CPU), no limit on multi-threading, and use at most 11/12 processors (in BOINC settings).