Hi,
I want to take control of all GPU related processes. To do this, Can we isolate GPU from all processes? So, we can give GPU access to our own processes (ex. a cuda program). Is it possible? Or do you have any idea to do similar thing.
Best regards.
NOTE: I’m using Jetson TX2.
see the accepted answer here
https://stackoverflow.com/questions/13900078/how-to-prevent-two-cuda-programs-from-interfering
the TCC driver on Windows will only support some Tesla and Quadro cards (maybe some high end Titans too). Note that while a GPU is in TCC mode, it would not support display capabilities.
Christian
Hi Christian,
I have already seen this answer. Because of Jetson TX2 restrictions, I can not use ‘nvcc’ unfortunately. Do you have different suggestion?
it may help to not run a graphical UI (X Server) on the TX2.
best to post this question in the dedicated Jetson TX2 forum.
[url]https://devtalk.nvidia.com/default/board/188/jetson-tx2/[/url]
Are you doing some latency critical processing that must not be pre-empted by other applications? Uunless you allow remote login by multiple users you are fully in control what programs will execute on the box. So why is there a need to use some kind of GPU exclusive mode?
Thank you @cbuchner1
Ok, I will post Jetson TX2 forum too.
Yes! Actually, I want to experience how safety critical deep learning systems, such as autonomous vehicles, should be run. So my model must produce stable FPS.
There is no remote user except me on Jetson TX2. But system processes such as GUI can access GPU. I want to isolate GPU from both users and processes. So I can not understand what do you mean.