cudaError 30 (unknown error) calling 'cudaMemcpyAsync(hash, context_hash[stream][thr_id], mem_size, cudaMemcpyDeviceToHost, context_streams[stream][thr_id])'
(D:/Christian/Documents/Visual Studio 2010/Projects/CudaMiner/sha256.cu line 446)
the "unknown error" after several minutes of successful mining might be a sign of a hardware related fault. Maybe something is unstable on the PCI express bus.
a better error recovery in CUDA would be nice to have, but it would not avoid running into these errors in the first place.
Try if the BIOS allows you to reduce PCI express bus speed (e.g. switching from PCI express 3.0 to 2.0) or to change some bus related options.
I've recently had a github support ticket open with the same error appearing with some Tesla boards. I've closed it somewhat disgruntled when the guy stated that I was about as helpful as a merry-go-round. Well, I cannot solve your hardware ailments, guys. I might work on implementing better error recovery though.
BTW: CUDA 6.0 shows much better error messages now (not just unknown error, or unspecified launch failure). But cudaminer has to be built with the CUDA 6.0 toolkit for this to work - and there's performance related problems with this.
Christian
I think its a bad board. IDK how it got bad but I moved it back to the "once working" rig and got the same errors. Unfortunately I just missed the 30-day exchange window at Best Buy by 4 days so I'm in talks with NVidia now hopefully get an RMA. I just don't know how much to tell these guys on the "how, why, where" etc. Im surprised there isn't an "anti-mining" clause in their warranty docs.
I saw that ticket on the GitHub site. That guy was being a D-bag!
We wanted to get mining fast.