I was working on Cygwin with my window ALT+ENTEREed (edit: corrected ATL_TAB to ALT_ENTER)
At some time, one of my colleagues ran a CUDA program (he was logged in via SSH).
When the program was launched, my entire screen was blacked out with strange colours. Although the keyboard controls were there, I could NOT do anything in the window. When I ALT+TAB back again, it was all normal and i had not lost any data…
My collague’s kernel reported no errors and validity check resulted in success.
So, what should I conclude?
Does my colleague’s kernel has out-of-bounds accesses?
Some other problem with display while CUDA is running? – i mean, is it a CUDA problem?
The behaviour is repeatable everytime. Our kernels have not nothing to do with OpenGL
The programming guide speaks about mode-switches, and how it can cause CUDA programs to fail. But it’s surprising that the CUDA program executed correctly and it was the screen that suffered, instead of the other way around…
Thats a very good point… Thanks for bringing this up. But the mode switch did NOT happen in the middle of kernel execution…
The mode switch happened and then only the CUDA program was started. So, the CUDA program could ideally fail in “cudaMalloc()” saying no memory.
But garbage on screen is too much… If u have windows system, can you just try with a sample kernel? Instead of cygwin command prompt, u could probably use the “cmd” prompt in ALT+ENTER fashion…