Jetson Nano L4T 32.5 CUDA 11.x Compatibility

Autonomous Machines Jetson & Embedded Systems Jetson Nano
1

Hi all,

No doubt a very simple question but I have not found a conclusive answer online. I have a Jetson Nano with the latest L4T 32.5 image loaded. It comes preloaded with CUDA 10.2.

I have installed CUDA 11.1 and built the samples but deviceQuery does not run.

CUDA Device Query (Runtime API) version (CUDART static linking)

cudaGetDeviceCount returned 3
-> initialization error
Result = FAIL

I can build and run the 10.2 deviceQuery sample which includes the following line:

CUDA Driver Version / Runtime Version          10.2 / 10.2

I assume this means base driver for the Tegra SOC does not support any CUDA greater than 10.2?

2

The CUDA toolkit won’t work without the CUDA driver of greater or equal version (which the L4T GPU driver is still on CUDA 10.2). Plus if you downloaded and manually installed the ARM64 SSBA version of CUDA Toolkit, that is for ARM servers using discrete GPUs (not Jetson, which uses an integrated GPU).

In short, use the version of CUDA toolkit that comes with JetPack, because it matches the driver version.