Hmm, I really don't know what's going on here. Could you try to launch Octane from the terminal (the executable is of course hidden inside the bundle)? Does it give any error messages?balt wrote:Here's the output of deviceQuery, the utility to find your CUDA capable devices that comes with the CUDA SDK. This confirms that there's no reason Octane doesn't find any CUDA devices...
Still no further ideas on what else to try from across the Tasman? Might this perhaps be the time to take this to a support email rather than a public forum? This obviously is a problem that is not affecting any other users at present.
I also checked which modules are being used by Octane and by the deviceQuery task that successfully sees my GPU. Interesting things I found there:
- Octane appears to make use of OpenCL as well as CUDA libraries (!).
- There appears to be one small difference in which libraries are used: Octane loads /usr/local/cuda/lib/libcuda.dylib, this is not loaded by the CUDA SDK samples. Conversely, libcudart.dylib is not loaded by Octane but used by the SDK samples. Perhaps there's a versioning problem there?
...
- CUDA framework, libcuda_295.10.20.dylib, libgpucomp.dylib are identical in both processes
Btw, the OpenCL part must have been pulled in by the CUDa driver (libcuda.dylib). And we don't use the CUDA runtime library.
Thanks,
Marus