Does current CUDA support SLI-mode multi-GPU programming?

[lyso1]

I've already known that CUDA can use { cudaSetDevice(0) & cudaSetDevice(1)} for multi-GPU task distribution,

but don't know whether CUDA supports SLI-mode multi-GPU programming.

If it does support, can the CUDA programmers just work on one logical GPU without knowing the number of physical graphics card in SLI mode?

[天下第一好大人]

For the beginning, CUDA processing conflicted with SLI-mode.

Now it is compatible with SLI-mode, but you must distribute your workflow yourself,
there's no logical GPU.

[lyso1]

[Quote=引用 1 楼 gogdizzy 的回复:]
but you must distribute your workflow yourself,there's no logical GPU.
[/Quote]

Cool, I'm just looking forward to this parlance!

Is there any nVidia-official documents showing this?

Thanks a lot!:)

[天下第一好大人]

I don't see any Official Docs but you can found a sample "CUDA openMP" in the SDK.

Generally, you can create multiple threads on the CPU, and each of these threads controls one CUDA device.

[OpenHero]

CUDA can't support SLI recently.
but you can use multi-GPU with multi-thread

[lyso1]

got it. thx a lot!

[lyso1]

test