Is it safe to use mixed driver and runtime

For various reason (Tensorflow etc) I have install Cuda-9.2 onto my Ubuntu 18.04 system. I had previously installed Cuda-10.0 using the .run package and not using apt-get. I also take the updates from Ubuntu for security reasons. Running nvidia-smi tells me that I have Driver Version 410.104 and Cuda Version 10.0. When I built and ran the Canny Edge Detector sample I get this:
GPU Device 0: “Quadro P5000” with compute capability 6.1

cudaSetDevice GPU0 = Quadro P5000
NPP Library Version 9.2.148
CUDA Driver Version: 10.0
CUDA Runtime Version: 9.2
Device 0: < Quadro P5000 >, Compute SM 6.1 detected

It seems to have run fine. So am I going to have problems down the road with a Driver version of 10.0 and a runtime of 9.2?


No, it’s fine. It’s part of a standard compatibility path in CUDA. Newer drivers work with older toolkits.