Hi!
I have been trying to build instant-ngp on a GeForce GTX Titan.
I have successfully build the application using cmake and the instructions from the github repository, but when running instant-ngp.exe I got the following error:
ERROR Uncaught exception: mypath\instant-ngp\dependencies\tiny-cuda-nn\include\tiny-cuda-nn/multi_stream.h:44 cudaStreamCreate(&m_stream) failed with error CUDA driver version is insufficient for CUDA runtime version
My Cuda toolkit version is 11.5 and my driver version is 456.71 (Cuda 11.1) which is supposed to be supported according to this doc.
Any idea of how I could fix this error? Thx a lot for the help!
Hello @jean.morell and welcome to the NVIDIA developer forums!
I saw you also posted this as an issue on the Github pages? Would be great if you could cross-post any answer you might receive there to help future viewers.
My first thought here would be that the actual CUDA driver might not include the feature set needed for the instant-ngp examples. That means while 11.5 runs on the 456.71 driver, it will do so with a limited feature set.
One possible solution would be to install the compatibility upgrade. Please check the documentation for details and follow it exactly if you try it out. Including the support matrix and further considerations.
Another thing to think about is the CUDA feature set. If you have a GTX Titan then it has still compute model version 3.5 (see here for more information). This might not be enough to actually run NeRF or similar. But for that you would need to ask on Github or in our CUDA forums.
I hope I could help a little!
Thanks!
Hi @MarkusHoHo ! Thanks for the answer and support. I didn’t get time for the moment to check all the docs that you sent, but will do it soon. I’ll get back here with the results, and of course if it works I’ll report what I’ve done here and on the github as well.