Upgrade cuDNN on Xavier

Hi,

I want to upgrade cuDNN libraries on Xavier. My current version is 7.5…0.56 and I want to upgrade to 7.6.3. . I checked the matrix page, my current CUDA version is 10.0 and above cuDNN version is supported. When I checked cuDNN downloads page
https://developer.nvidia.com/rdp/cudnn-download , I see download options for x86_64 and power architectures. I don’t see any links for ARM.

Where can I find cuDNN v7.6.3 installation files for Xavier. (and also for Nano).

Hi,

You can use Jetpack 4.3 SDK using SDK manager

Thanks

Thanks for your reply, I don’t want to re-flash my Xavier to just update one library. Is there anywhere I can just download compiled library like what you have for x86_64 archs.

Also, I believe I am on Jetpack 4.4

$ head /etc/nv_tegra_release

R32 (release), REVISION: 2.3, GCID: 17644089, BOARD: t186ref, EABI: aarch64, DATE: Tue Nov 5 21:48:17 UTC 2019

In order to update TRT/cuDNN version, you should use SDK manager only.

Thanks

OK. If I use SDK manager, can I selectively update, instead of re-flashing Xavier.

I just checked and I am already on Jetpack 4.3SDK and still my cuDNN version is 7.5.0
dpkg --list | grep libcudnn
ii libcudnn7 7.5.0.56-1+cuda10.0 arm64 cuDNN runtime libraries
ii libcudnn7-dev 7.5.0.56-1+cuda10.0 arm64 cuDNN development libraries and headers
ii libcudnn7-doc 7.5.0.56-1+cuda10.0 arm64 cuDNN documents and samples

How can i upgrade to 7.6. Can one of you point me to source and I can compile it.

Hi Nvidia team, any help on this?

Can someone point me to source and I can d/l it and compile it on Xavier.

Hi,

Sorry for late reply.
Jetpack 4.3 SDK has TRT 6 and cuDNN 7.6.3
Please refer to below link:

Also you can select specific section while flashing, please refer below link:

Thanks