I have created project in C++ for CUDA GPU. it runs OK, I tried for 1 000 000 paths for 512 european options, then I increased the number so loop took longer and longer (but still few seconds), but when I turned loop to really huge number program started, but during computation my Ubuntu 12.10 started to turn off. I turned off my comp manually, because I don't want my GPU tu burn off or something similar.
I have 4 cores amd 64, 8 GB RAM. may it be that GPU started to require too much power than possible and system decided to shut down? it was not a stack overflow, also I have seen RAM on machine wasn't involved. do I have limitations when using CUDA? can I have broken it?
Now I do computation on GPU again but the question remains: what are limitations, do I have to be careful about something?