CUDA version for tinycudann using CUDA_ARCHITECTURES=90
Anji-Builds opened this issue · comments
I need to use the tiny-cuda-nn to build a docker for Emernerf repo. I need to use a cluster containing h100(CUDA_ARCHITECTURES=90). When I checked the product specification of the GPU, it requires min. CUDA 12.x and also here
However, I noticed that tinycudann is recomended to work on PyTorch 2.1.2 with CUDA 11.8. So, wouldn't this lead to a conflict or mismatch?
Is specifying TCNN_CUDA_ARCHITECTURES =90 enough? Else, what changes do I need to make to successfully install tinycudann and get it working?
I am very confused how to go about it and I am really new to this especially building dockers.
Looking forward to some solution or guidance as early as possible.
Thanks in advance.