Hello so I am a new user for JetsonTx2 and I just installed the software and everything went smooth. However, I am trying to run a sample on it but I realized I need a monitor and a keyboard attached to the board. I do not have either, but is there a way that I can use the same computer in which I installed the software as the monitor and keyboard for the Jetson TX2? Please let me know step by step how and what to do since I am a newbie and dont know much about it.
Thanks in advance.
If you have configured the IP address of your jetson, you may ssh into it.
You may look also to -Y option for ssh.
One thing about “ssh -Y” is that some of the GPU function will offload to the PC…you might have a surprise. A virtual desktop would pretend there is a monitor and keyboard attached, and forward results to a PC.
Hi Linuxdev,
Thanks again for adding this useful information.
I would assume this to be relevant only if the ssh comes from a CUDA-enabled host. Could it be otherwise ?
Furthermore, is there a compute capability check ? For example, if I ssh -Y from a TK1 into a TX2, would it try to offload TX2 by loading the TK1 ? For X display, local transforms may be better, but for CUDA applications, it is not so obvious, IMHO.
If you require CUDA of a specific version on the Jetson, then offloading to host will also require this on host…the CUDA of the Jetson would be abandoned and the PC would take over. Similar for OpenGL and OpenGLES…the specific version the program on the Jetson requires would instead be required on the host, and the host would provide the hardware. Using a virtual desktop would allow all requirements to stay on the Jetson (the PC host would simply watch what goes on instead of participating). Note that a TK1 uses 32-bit CUDA and TX2 uses 64-bit CUDA…the two would have an issue doing “ssh -Y” remote display from one to the other.
Ok, I won’t try to ssh -Y from TK1 to a aarch64 jetson ;-)
If I ssh -Y from a TX1 to a TX2, where I run a program compiled for cuda-8.0 with compute capability 6.2, could it be offloaded to TX1 if it has cuda-8.0 but has only 5.3 compute capability ?
That’s a good question…I can’t answer. I’ve seen another thread where the CUDA compile for Pascal failed to be interchangeable with Maxwell…but the program could be configured for both GPU architectures (it was an advanced setting in nsight eclipse). I suspect the compute architecture must match.
Do you guys know if there are any videos to guide me through the ssh set up?
SSH is a fairly standard setup within the Linux world…it can get complicated if doing something special, or if using Windows via something such as PuTTY. Can you be more specific about your particular case which you want to work with, e.g., between an Ubuntu host (or other Linux flavor) and Jetson, or perhaps you want something more sophisticated like ssh keys?