Quote from: ltcnim on March 28, 2014, 12:25:52 AM

Quote from: cbuchner1 on March 28, 2014, 12:07:37 AM

Quote from: ltcnim on March 27, 2014, 11:06:54 AM

Quote from: cbuchner1 on March 27, 2014, 07:56:21 AM

Quote from: ltcnim on March 27, 2014, 01:32:49 AM

5 x eVGA 750Ti SC non overclocked = 61MH

but I'm unsure how to explicitly compile it for sm_35 and 128 regcount this time (due to the new Makefile.am). Everytime I get invalid calculations for the shares. But I think compiling with standard settings is a bit slower on groestlcoin atm. Any advice?

on compute 3.0 and above we're using a block size of 768 which cannot work with more than 80 regs per thread.
it's the reason the 750Ti jumped from 9 MHash to 11 MHash... just like that Wink

ah, thanks for clarification. just two more questions:

1. under linux, after compiling all compute settings are within one executable? how can I make sure, that sm_35 is used?
2. is there a possibility to change regcount for a specific algo (in my case groestl) while having it untouched for all the others?

unser Linux it uses the compute setting matching your card best.

add a file specific rule to Makefile.am vor groestlcoin kernel. Nur IMHO you should try upping the groestl Thread block size to 768 first...

hmm..is that the value in cuda_groestlcoin.cu:

const int threadsperblock = 512;

? or do i need to change it somewhere else?