Below are the values I am using in the gpu.config for the new sieve tasks.
I haven't had much time to test it, but it is loading my 7970s at 99%. It looks like with this gpu.config I am doing about 2.5 Million PPD on each of my 7970s.
I tend to run mine at
kernels_per_reduction; is the most critical one, most AMD cards can easily support a setting of 50 here higher end 52.
threads; again anything 7950 and above can support 8
sieve_size; sets the amount of work each cycle, 27 is good 28 is better will also work.
kernals is the critical one. on mine, 7970's & 7990's, they easily run at 54, anymore and the processing starts slowing down. 2 minute work unit is the target. (although I have had them down at 1:45)
Oh, and btw:
I've joined the team on collatz
Edited by Egilman - 1/3/16 at 12:14pm