Jetson Nano Pytorch returns CUDA False


I just checked your full log again and figure out what is going on

  1. You are actually using a custom board from specific vendor with a emmc module. (No matter what the vendor said, they are not real NVIDIA devkit)

NVIDIA provides 2 kinds of modules. One kind has emmc but no sdcard slot. Another kind is with sdcard slot but no emmc. Sdcard module only sold along with NVIDIA devkit. The sdcard slot will be on the “module” but will not be on the carrier board.

The so-called “sdcard image” can only work on sdcard module. If you directly use any kind of BSP provided by NV, then it won’t be able enable the sdcard slot you have there, because your sdcard slot is add by the vendor and on the carrier board but not the module.
NVIDIA image does not enable such thing.

To enable such sdcard slot, you need to consult with the board vendor and they shall provide you the customized bsp.

  1. Your emmc already has a new kernel version “4.9.299-tegra”. But the file system, which is loaded from the sdcard (actually a usb device to jetson) are using a old version of jetpack.

In brief, your emmc is probably installed jetpack4.6.4 but your sd/usb still uses a 4.6.1 image. This causes linux kernel image version does not match to the linux kernel modules (then nvgpu driver is not loaded).

Remove your sdcard there and it shall totally boot up from emmc and everything shall be matched there. (if you didn’t install emmc correctly, then it may not work either).

It appears the emmc has not been installed correctly. How could I modify your procedure to install a jetson nano image on a device with emmc? Is that something that’s possible?


I really don’t know how the vendor taught when you get your jetson nano on the first day. Or they totally gave no tutorial.

There is a tool called sdkmanager that can flash jetson BSP (jetapck) from another x86 host to jetson.

This is the basic tool that you should learn from day 1. Sdcard image does not really flash anything.

Also, just to mention this again. You are using a custom board and we don’t guarantee our software can 100% support all the functions on your board. You need to check with the board vendor to get their customized BSP.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.