CUDA 10.2 nvidia-drivers 450.51 needed for Tesla V100 to install DS5.0 but only 440 available?

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

Hi All,

I was previously running DS4.0 with Nvidia GPU Driver 418, CUDA 10.1 and TensorRT 6.1 with Tesla V100 & Ubuntu 18.04.

DS5.0 update guide https://docs.nvidia.com/metropolis/deepstream/dev-guide/text/DS_Quickstart.html#dgpu-setup-for-ubuntu mentions 450.51, CUDA 10.2 and TensorRT 7 is required to install DS5.0.

However resources page to download Nvidia GPU drivers: https://www.nvidia.com/en-us/drivers/unix/linux-amd64-display-archive/ links only to 450.51 that does not seem to be compatible with Teslas https://www.nvidia.com/Download/driverResults.aspx/160555/en-us
I searched here https://www.nvidia.com/Download/Find.aspx?lang=en-us but for CUDA 10.2 there are only 440 drivers available.

Please advise how should I proceed to properly upgrade to DS5.0

3

Hi,
Display driver 450.51 is for cuda 11.0, you can download it and install the driver, then install cuda toolkit 10.2, please go ahead and follow the document to upgrade.
any follow on issues raised please let me know.

4

Hi,
I installed 450.51.06 and CUDA 10.2 as advised and looks like everything works fine. I just see there is a mismatch:
nvidia-smi gets me:
±----------------------------------------------------------------------------+
| NVIDIA-SMI 450.51.06 Driver Version: 450.51.06 CUDA Version: 11.0 |
|-------------------------------±---------------------±---------------------+
but
$ cat /usr/local/cuda/version.txt tells me CUDA Version 10.2.89

What is the meaning of CUDA Version as reported by nvidia-smi?

5

The nvidia-smi tool gets installed by the GPU driver installer, and generally has the GPU driver in view, not anything installed by the CUDA toolkit installer. the (installed) driver API version may not always match the (installed) runtime API version, especially if you install a GPU driver independently from installing CUDA (i.e. the CUDA toolkit).