Cross compile TX1 with a PC with GTX 1080

I’m having an issue to run the example on TX1. I get the following error: “CUDA driver version is insufficient for CUDA runtime version”.
My dev machine has Cuda 8.0.61 while the TX1 has 8.0.33. I checked my GPU and PTX for 5.3 generation and still the same problem. I can compile and the samples normally directly on TX1, but the cross compile isnt working.

So I figured out I can’t (correct me if I’m wrong, please) compile local projects with Cuda 8.0.33 on my Dev machice coz I have a GTX 1080 GPU and it doesnt work with nvidia-361, which works with cuda 8.0.33. I even manage to use cuda 8.0.33 to cross compile, but then I can’t compile local projects.

How can I align cuda versions to cross compile and local compile in this scenario?

Thank you so much


Thanks for your question.
Do you use Nsight for cross-compiling?

We tested Nsight to cross-compile the vectorAdd sample and it works well.(We also have different cuda version between host and device).

Please make sure you have select “AArch64” CPU Architecture in Target Systems session.

Please let us know your results.

Hello @AastaLLL, I am have the same problem that was described above except my host does not have a GPU. Set is as follows:

Host PC:
Intel i7-6700 CPU @ 3.40GHz
Ubuntu 16.04.4
Cuda Toolkit version 8.0.61

Target PC:
Jetson TX1 board Aarch64
Cuda NVCC version 8.0.33

I tried your solution and made sure I selected AArch64 as the CPU architecture in the target systems session in Nsight however the error still has not changed. I have tried on both the boxFilter and vectorAdd samples and both of them state (error code CUDA driver version is insufficient for CUDA runtime version)!. Please let me know if there is any other action I may be able to take to fix this issue.

Hi ncooper,

Thanks for your feedback.
Could you follow this page and try it again?

We have verified this procedure and Nsight work properly.

Please give it a try and let us know the results.