Would I benefit from second GPU with my system

I was reading about how people speed up the blender interface by setting their graphics card to the integrated graphics card, and thus using their GPU just for rendering. And the intergrated card for the interface and stuff. Which apparently…makes a difference in blender performance, because the GPU doesnt have to worry about running the programme anymore and drawing all the monitors.

This might be something i want to try, I have a 3 monitor setup and blender interface is lagging quite a bit somethimes… So I want to see if this solution works for me. However… i dont have a intergrated grapics card on my motherboard. However, I do have an extra GPU lying around. A GTX 970.

My question is:
Would this also work in making the programme faster? Not rendering, but Blender itself? If I dedicate that 970 to just the programma…and use the 1080 for the viewport/rendering?

If so, how would I set it up in the Nvidia control panel?


  • GTX1080i and GTX970
  • Threadripper 2950x
  • 64RAM
  • Designare gigabyte x399 mobo
  • windows 10 64 bit
  • Blender 2.79

No, it would not be faster outside rendering. It would be slower, because the 970 is slower. Blender doesn’t use more than one GPU besides for Cycles.

So you can tell blender to use the integrated card to use for the programme and your screens… but not another GPU to do this? That sounds weird to me

Why can I do this with the onboard graphics card …but not with a second GPU? Isnt the principle the same?

Of course you can assign your 970 to the display. You just asked if your display would be slower or faster. Your display would be slower on the 970 than using your 1080ti for the display. But your rendering would then be slightly faster.

People tend to use the onboard graphics for this because it consumes the least electricity and doesn’t take up PCI slots they could use for extra rendering GPUs.

It is not a speed up itself.
When you use GPU connected to your monitor as the primary display for rendering with Cycles, this GPU takes so many resources that it makes your entire system slow while render. So “so far” the rule was to use a second not so powerful GPU to handle the display, and more powerful GPUs to render with Cycles.
I said “so far” because in 2.8 we will have Eevee and it is likely that you want to have the most powerful GPU connected to your monitor as primary display to get the best performance.

Yes this is what I meant… using a less powerfull display to handle the display.
But i would only benefit from this…when Im rendering? So that the rest of my system wouldnt slow down?

So this approach would not be helpfull if Im not rendering? To speed up the interface?

That’s not the whole story though. With the new driver model in Windows 10, for GPUs that are used for display, MS forces something on that used to be off, and it makes CUDA somewhat slower. I can’t remember what the details were or find the relevant conversation, but generally under windows now, any GPU used for display is slower for CUDA. It still works fine in Linux.

Yes, only while you render with Cycles. For example, you could use your machine to work on another project while Blender render with Cycles. This is something that you could not probably do if you use to render the GPU connected to your monitor.

probably this?

1 Like

Ah ok I get it… It seems that in my case …it wouldnt really make blender any faster (the programme…not rendering). So ill leave it out.

Thanks everybody for their replies

Yup, that. Hadn’t realized it was only for Quadros and Titans.