Hello All,
Another novice question.
I have a GTX 460 on my desktop and am getting familiarized with GPU programming.
My question is can i run two or more different kernels doing different jobs at the same time in parallel ?
For e.g. let kernel A does addition and kernel B does multiplication of two arrays. I want kernel A to use lets say 10 cores and kernel B to use remaining cores on the GPU simultaneously.
Is this possible? If so please let me know some pointers on this.
Thank you,
Jacky,
Hello All,
Another novice question.
I have a GTX 460 on my desktop and am getting familiarized with GPU programming.
My question is can i run two or more different kernels doing different jobs at the same time in parallel ?
For e.g. let kernel A does addition and kernel B does multiplication of two arrays. I want kernel A to use lets say 10 cores and kernel B to use remaining cores on the GPU simultaneously.
Is this possible? If so please let me know some pointers on this.
Thank you,
Jacky,
Yes, the Fermi GPUs like the GTX460 can run multiple kernels at one time. You do NOT specify which MPs run which kernel, though. The GPU itself will microschedule them more efficiently on a block by block basis.
In order to run more than one kernel at once, they both must be launched from the same CUDA context in different streams. This is discussed in the CUDA programming guide in the CUDA toolkit… look at the stream management discussion.
Yes, the Fermi GPUs like the GTX460 can run multiple kernels at one time. You do NOT specify which MPs run which kernel, though. The GPU itself will microschedule them more efficiently on a block by block basis.
In order to run more than one kernel at once, they both must be launched from the same CUDA context in different streams. This is discussed in the CUDA programming guide in the CUDA toolkit… look at the stream management discussion.
Thanks much for the reply. Exactly what i was looking for.
Thanks much for the reply. Exactly what i was looking for.