Running Jetson TX2

Autonomous Machines Jetson & Embedded Systems Jetson TX2
1

Hi, I have a CPU that going to be a host for Jetson TX2. I have installed Ubuntu 16.04 and Jetpack 3.0 in that Host, As I follow in GitHub - dusty-nv/jetson-inference: Hello AI World guide to deploying deep-learning inference networks and deep vision primitives with TensorRT and NVIDIA Jetson..

But after I tried to Install Nvidia driver, I cant find it when I list it on lsmod, I think its because my host is just a CPU, not CPU. When I tried to run ‘sudo make’ with cuda, it also showed some errors.

My question, How to run jetson TX2 from another Host which is not GPU, and just a normal CPU? is that possible? thank you

2

To run or compile CUDA packages on a system it must have the NVIDIA video card and driver. Running or installing CUDA software on the Jetson does not require the host to have any GPU or CUDA install…though it is a covenience in most cases. When compiling on the host for a cross-compile to a foreign architecture this should not matter (the TX2 is arm64/aarch64/ARMv8 foreign architecture to a host). Attempting to cross-compile without proper setup would probably fool the compiler into believing you are compiling for the host’s native architecture, which would be incorrect. I’m not sure if JetPack can ignore a host without a native CUDA/GPU-capable driver, so a question occurs…are you using JetPack to build for a Jetson and JetPack has issues with this, or are you compiling on command line and not via JetPack?

NOTE: On a PC the Nouveau driver is a software-only driver not supporting direct GPU hardware access. For a host to use CUDA Nouveau must be blacklisted and the NVIDIA direct hardware access driver installed. How to do this manually depends on Linux distribution.

3

Hi,

Nvidia driver is for desktop GPU. For example, if you want to use DIGITs for training.
If you want to use tx2 GPU only, the driver is contained in JetPack OS. No extra driver is needed.

4

Hi,

Nvidia driver is for desktop GPU. For example, if you want to use DIGITs for training.
If you want to use tx2 GPU only, the driver is contained in JetPack OS. No extra driver is needed.

5

Hi linuxdev, because Im still new to Linux, so maybe I did it wrong. When I compiled it, I compiled it via command line (terminal) not via jetson. Would that be wrong? should I compile it in Jetpack?

Hi AastaLLL, I only want to use tx2 GPU only, Is there any tutorial to only use tx2 GPU only without a GPU host? Because I cant find any example out there.

thank you

6

Hi,

There are two approaches:

  1. Write your program directly on the device.(via ssh or connect a display to the TX2)

  2. Use cross-compiler NsightEE:
    https://devblogs.nvidia.com/parallelforall/cuda-jetson-nvidia-nsight-eclipse-edition/

7

Command line compile won’t change things versus from a GUI/helper…the question is which computer the tool or command line is logged in to…Jetson or PC. Some tools (such as nsight eclipse, and to some extent JetPack) are capable of working on either of those environments simultaneously, which can be confusing. It is perfectly reasonable to compile on Jetson while using ssh from host, or directly logged in to the Jetson; it is also perfectly reasonable to cross compile on the PC host with a target of PC or a target of Jetson…or even to ssh from PC host to Jetson and compile there for Jetson.