We have an applications that runs fine on a 8800 Ultra, but when running it on a FX360M or 8600M GT sometimes a kernel launch error occurs or the computed results are wrong.
Its not really deterministic, when which error occurs … it also depends of the problem size and the number of threads used in a CUDA block.
Anyone has an idea what’s going wrong?
Sorry, but at the moment we can’t post a code. We didn’t really know which kernel produces the error, when testing each kernel independently the kernel works quite fine.
Thanks and regards,