Can cudnn8 run on tesla card 410, 396 driver with cuda10.2 docker?

AI & Data Science Deep Learning (Training & Inference) cuDNN
1

docker nvidia/cuda:10.2-cudnn7-devel-ubuntu18.04 supports 396, 410 …
ENV NVIDIA_REQUIRE_CUDA=cuda>=10.2 brand=tesla,driver>=396,driver<397 brand=tesla,driver>=410,driver<411 brand=tesla,driver>=418,driver<419 brand=tesla,driver>=440,driver<441
while in docker nvidia/cuda:10.2-cudnn8-devel-ubuntu18.04
ENV NVIDIA_REQUIRE_CUDA=cuda>=10.2 brand=tesla,driver>=418,driver<419
does this mean cudnn8 can’t run on 396 410 driver?
in cudnn docs, i didn’t see any driver requirements, there’s only cuda version requirements
i m also wondering if the nvidia/cuda:10.2-cudnn7 docker support driver 384.111+
because triton-server:20.03.1 support driver 384.111+ and it is base on cuda 10.2

2

Hi @763730968
please refer to the below link -

where , For the CUDA driver requirements for a given CUDA Toolkit version, refer to the CUDA Compatibility documentation.

3

thanks,
but i want to know the supported versions NVIDIA CUDA, the CUDA driver, and the hardware for the NVIDIA cuDNN 8.0.0 not 8.8.0
and cuda10.2.89 with cudnn7.2.1

4

Hi @763730968 ,
If I understood the query correct, you can find them here.

Thanks