Upgrade CUDA libraries on TX2 ?

We have a TX2 system which was originally flashed with Jetpack 3.1. The system has been intensively customized since then and we just cannot do a fresh install of Jetpack 3.2.1.

However, we would like to try new CUDA 9.x (we have an older CUDA 8.0.84 installed right now). How can we install a new CUDA release? Does it have dependencies on other components? If anyone already did it and can share their experience, it would be greatly appreciated!

Also, what about cuDNN and TensorRT? Can the new releases be installed side-by-side on a Jetpack 3.1 based system?

-albertr

Hi albertr, it isn’t officially supported because CUDA9 relies on updates in the BSP which is tied to the JetPack version.

I think your time would be better spent by cloning/backing up your existing image and saving out your components for re-install.

Dustin, thanks! I’ll be fine without having “official” support ;-) Reinstalling OS is not an option (I’m sure alot of developers do agree on this). It would be nice if NVidia would continue to support existing customers who are using Jetpack 3.1 (less than a year old, BTW).

-albertr

Re-flashing with new JetPack (if you want the upgraded components) has always been the Tegra/Jetson modus operandi and is typical for embedded development under the host/target paradigm. Due to the interdependencies of the platform, it’s unrealistic to issue upgrades to previous JetPack’s other than critical security fixes (which don’t include major component updates).