The CPU problem is fixed but there's still a problem. I have a linux rig with a 970 and a 780ti, compiled
ccminer for compute versions 3.5, 5.0 & 5.2.
When I start mining neoscrypt on both GPUs I get nothing but HW errors on the 780ti (GPU1). But I can mine
each GPU in seperate instances of ccminer.
This fork doesn't support compute 3.5. Maxwell only. (5.0 and up). but it might work for some algos. Don't have a 3.5 card anymore.
I don't know what he hardware errors were about but they've gone away.
Neoscrypt vs r58: 970 550kh (+25k), 780ti 335kh (-45k), +120 OC on both.
It looks like I'll be keeping r58 for mining neoscrypt on the 780ti.
I have a working prototype of r59 with the best of both worlds. It gets the new hash rate on maxwell
and the old hash rate on kepler. I just copied the old neoscrypt kernel from r58 and gave it a
new name. The GPU architecture is checked in each thread to decide which kernel to use.
I chose to do a full copy of the r58 (kepler) kernel so that any future changes to maxwell code won't break
kepler. It was also the quickest and easiest implementation. There's a lot of duplicate code as a result.
It could probably be streamlined but with no future support for kepler expected it's probably better
to keep it as a self contained unit.
I put the gpu check in ccminer.cpp because it was easier but it might be a better design to move it
lower into the neoscrypt code.
I can make the changes available if any of the ccminer devs are interested in reviewing them or
including them in their fork.
