I’m attempting to get CUDA up and running on OS X 10.12, on a MacPro 6,1 on OS X 10.12.2, using a 960 in a Powercooler Devil Box. I believe the machine recognizes the GPU, as I was able to get the drivers installed.
The installed drivers are GPU Driver - 367.15.10.25f01 and CUDA Driver 8.0.57, both which claim to support the current version of OS X.
I’ve followed the OS X CUDA Set up guide through to the verification step, and am trying to run the deviceQuery, and get the following output
mac-pro:release thiney$ pwd
/Developer/NVIDIA/CUDA-8.0/samples/bin/x86_64/darwin/release
mac-pro:release thiney$ ./deviceQuery
./deviceQuery Starting...
CUDA Device Query (Runtime API) version (CUDART static linking)
cudaGetDeviceCount returned 35
-> CUDA driver version is insufficient for CUDA runtime version
Result = FAIL
Unless I’m reading something incorrectly, these drivers should be compatible. Any idea for what to do?
Where did you get this driver: 367.15.10.25f01
I don’t think those are the same drivers, as one ends in 25f01 and the other on the NVIDIA web site ends in 25f02
Also, when I look at the supported GPUs tab on the NVIDIA web site, the 960 is not listed, nor is macpro 6,1
The 25f01 driver seems to be here:
[url]http://www.nvidia.com/download/driverResults.aspx/113889/en-us[/url]
but it doesn’t list your config as supported either.
I also note that there is an instruction there (on either the 25f01 and/or 25f02 driver posting):
“In order to run macOS Applications that leverage the CUDA architecture of certain NVIDIA graphics cards, users will need to download and install a compatible CUDA driver. CUDA driver which supports macOS Sierra 10.12.2 will be available at a later date.”
Since that entry is dated in 2017, and the last available Mac CUDA driver posted at the moment is:
[url]http://www.nvidia.com/object/macosx-cuda-8.0.57-driver.html[/url]
which is dated 12/15/2016, it seems like there may be an issue there as well.
The only thing I can suggest is maybe give the 25f02 driver a try.