Run CUDA 10.1 on Jetson Nano

Robotics & Edge Computing Jetson & Embedded Systems Jetson Nano
1

There is any way to run CUDA 10.1 on Jetson Nano?

I was trying to update from Jetpack 4.2 but there aren’t any arm64 compatible packages, only x86 and PPC.

I’m running Linux 18.04 from the latest Jetpack currently available.

Thanks,

2

Hi ghostvmg,

We only provide CUDA installation through JetPack, that’s due to the corresponding/dependencies in platform libraries/drivers. We will support CUDA 10.1 on Jetson platform soon, please stay tuned.

3

Hi kayccc,

Thanks for your info.

I will wait for it.

Do you have any date planned?

Thanks,

4

I would also be interested in when it’s going to be released, as currently there’s a mismatch of PyTorch ONNX-opset for export and onnx2trt opset for building TensorRT-models, which is quite annoying (and fixed when building against tRT 6.0 on desktop)

6

Any progress? Are you guys just playing golf?

Terveisin, Markus

7

You can see the core component versions in the upcoming JetPack 4.3 release here: https://developer.nvidia.com/jetpack-4_3_DP

JetPack 4.3 includes TensorRT 6.0 and should be released in the coming weeks. The next CUDA upgrade to JetPack is planned for the spring.

8

Thank’s for update, Markus

9

Is there already a more detailed release plan for the JetPack version >4.3 ?

I really would like to get TensorFlow 2.1 to work on the Jetson Nano, since it does support TensorRT natively, but it requires CUDA 10.1. Since a lot of people are using TensorFlow on the Jetson Nano and they would highly profit from TF 2.1 TensorRT support, it might be worth to bump up the priority of upgrading the JetPack CUDA version to >=10.1 :)

10

Hi Langhalsdino, yes the next JetPack release will upgrade the CUDA toolkit version as well.

11

Did you successfully install Tensorflow-gpu2.1 on Jetson nano?