install CUDA on a brand-new TX2?

Robotics & Edge Computing Jetson & Embedded Systems Jetson TX2
1

Hi,

My hardware guys just unpacked a new TX2, we booted it, signed in, ran the startup stuff (NVIDIA_INSTALLER) and… no CUDA. No /usr/local/cuda, no nvcc (anywhere - with a “find / -name nvcc -print”), just some libraries. I’d just as soon not reflash the thing if I don’t have to, but I can’t seem to find a downloadable installer (.deb) of just cuda anywhere (developer.download.nvidia.com, from some online help, is “404 not found” for everything). The Jetpack 3.1 (which matches my Tegra release) installer files, when unpacked, do not contain much except “cuda-lt4.sh”, which wants to do an apt-get, but “apt-cache search cuda-toolkit” on the TX2 itself returns nothing at all (after a “apt-get update”.)

Here’s /etc/nv_tegra_release, if that helps. Ubuntu is 16.04

R28 (release), REVISION: 1.0, GCID: 9436269, BOARD: t186ref, EABI: aarch64, DATE: Fri Jul 28 17:04:29 UTC 2017

Any suggestions? Thanks!

2

JetPack is what provides the CUDA toolkit for aarch64 architecture and Jetson. Since you have an outdated version of L4T (the latest for TX2 is R28.2.1), it is recommend to re-flash with JetPack. You can have JetPack install CUDA, cuDNN, TensorRT, ect. for you after the flashing step. JetPack runs from the host PC, with you Jetson connected to the host over micro-USB and ethernet.

3

OK - if that’s the way, that’s the way! I’ve refreshed our TX1’s recently, so I kind of even remember how to do it :-)

Thanks for the answer!