Full Subject: cudaMemcpy CUDAStream::Download failed unspecified launch failure
I have a Fedora 10 system running with a GF9400GT that has been running stably for some time.
I'm using version 177.67 of the nvidia drivers. nvclock keeps the fan speed cranked up to 80%
for me just fine.
Recently, I'm getting the above error--seems to have started about 8 hours ago. I'm also getting
new work units that I've never seen before: 660x Starting with these, I get this error and the machine
quickly goes into 24 hour wait due to EUE.
Anyone else seen this? This appears to be a Core_11 job (version 1.31 here).
The jobs fail immediately after the "Working on Protein" line, so there's little chance these are
thermal problems for the card--especially considering it's a 9400 which couldn't get hot if it wanted
Yes, I see you all quaking in fear at the awsome power of 16 shaders.
Further investigation: Another machine I have is running F12 and uses the newer beta driver version
195.17. He has a GF9600GSO (old 96 shader type). The same work units work just fine on him.
Could this be a wrapper/driver version issue?