double C870 tesla in SLI SLI mode doesn't work...

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,

I have 2 C870 Teslas in SLI (2 * PCIe 16x and the SLI link between the Teslas) with a NVS280 for the display.

When I run “deviceQuery”, the 3 cards are detected, instead of two (the NVS and the “double C870”).

I have no menu in the nvidia control panel to activate the SLI mode, and curiously, the IRQs of my NVS and of one Tesla are 0 whereas the second Tesla IRQ is 16…

:blink:

For information, I have about 100 fps with the original size window in “fluidsGL” program.

Thx

2

And why do you need SLI? If you manage to enable SLI somehow then CUDA will see only one Tesla and will use only ‘half’ of SLIed device. Your current config is correct, IMO.

3

SLI is for graphics applications only. CUDA can access multiple GPUs without the need for the SLI connector or changing the control panel.

4

I don’ t think so…

2 cards connected in SLI made a virtually one with double block capacity (better for a generic program)

2 GPU without an SLI link have to be managed separatly…

H.

5

SLI is not supported in CUDA. SLI does not make a device with double block capacity. It only connects framebuffers as I understand it.

6

Ok

Thx for all your answers. :biggrin:

H.