Geek asks and answers

Development for CUDA on "cheap" GPUs

I am developing algorithms in CUDA on my desktop that will later run on the server.

Is it possible to use a recent low-level card (for example, computational ability 2.1) to get all the good debugging and profiling functions, and then put the code on the server using a high end card (with the same cc)? I just need to adjust the thread / mesh sizes or change everything ™.

Example: I would develop a Quadro 600 , and the server would use the Tesla C2075 .

+3
source share
3 answers

Until your kernel call and the scalable kernel itself have problems.

Check out this question:

CUDA ?

+2

, (25,6 / Quadro 148 / Tesla, ) SM ( SMs -). .

+1

, Multi-GPU, . , .

; , , , GPU, SM.

+1

More articles:


All Articles