GTX1070ti + GT1030 for multiple displays

Hi all,
I have bought gt1030 near by the main GPU 1070ti

I need to set the gt1030 only for the main display
And the 1070ti is for main render, physics and everything

I can’t do it,
When i connect the main display for the gt1030, i found the PC Use it as main GPU for render

I need to use it for main display only,

And the 1070ti is for multiple display outputs with render and everything else

Please support

In windows, for general graphics applications, its not possible to force the render to be on one GPU while the display is on another.

It’s not clear why you are concerned about this.

PhysX can be assigned to a particular GPU; use the NVIDIA windows display control panel to do this.

CUDA can be assigned to a particular GPU; use the CUDA_VISIBLE_DEVICES variable.

Speculation: OP might be concerned that the GT1030 is not fast enough for the rendering task.

Generally speaking, for CUDA work that requires a minimum of graphics output (or none), choosing a fast GPU for compute and a cheap low-end GPU to drive the display (OS GUI, etc) is fine.

However, if a CUDA accelerated application also has complex rendering requirements (e.g. molecule visualization), one may need a fast GPU for that, too. I seem to recall that in the past, people deployed workstations equipped with a Tesla/Quadro combo for such use cases. That was when there were still actively-cooled Teslas for workstations; no idea what people deploy for these use cases today.

The usual suggestion for workstation compute would be multiple Quadro cards. On windows, most Quadros can be placed in TCC mode. Place your compute Quadro in TCC mode. On linux, simply don’t extend your X-server to the compute Quadro.

Graphics, vis, interop would all be directed (automatically) to the display Quadro. CUDA compute tasks would be directed to the compute Quadro.

Thanks, that’s good to know for future reference.

Thanks dear for your reply,

I did the step for PhysX as you say

But the CUDA i can’t find CUDA_VISIBLE_DEVICES in Nvidia sittings

Where to find it?

CUDA_VISIBLE_DEVICES is an environment variable. If you are on Windows and don’t know what that is: