Network node RT - GPU rendering performance?

[ripe]

Quick question.

I'm using RT more and have 2 x 780ti GPU's in my main PC.

What's the performance hit of using distributed RT rendering with a 2nd PC, so for example if my 2nd render node PC had 2 x 780ti GPU's in it also, should I see roughly twice the performance by bringing it into the mix - or does the network overhead etc mean it's a lot slower than having all 4 cards in my main PC.

Main reason for asking - is if I want to add more GPU's to my main machine I will need a new motherboard - but I can easily network more if it's worthwhile?

[Morbid Angel]

RT renders in progressive mode on gpu, the main bottleneck will be the network speed. For this reason I still use bucket and cpu.

[biochemical_animations]

I have a current project, where a 1280x720 renders on my WS with dual 980 ti's at 9.5 minutes. I can DR render across my render farm with 5 nodes, each with dual 980 ti's and only cut the render time in half to 4.25 minutes.

So that is 2 cards at 9.5 minutes or...
12 cards at 4.25 minutes.

This is for RT CUDA...DR helps but I think the best might be a WS with multiple cards. I'm in the process of building such a WS.

[Dariusz Makowski (Dadal)]

I have a current project, where a 1280x720 renders on my WS with dual 980 ti's at 9.5 minutes. I can DR render across my render farm with 5 nodes, each with dual 980 ti's and only cut the render time in half to 4.25 minutes.

So that is 2 cards at 9.5 minutes or...
12 cards at 4.25 minutes.

This is for RT CUDA...DR helps but I think the best might be a WS with multiple cards. I'm in the process of building such a WS.

This don't look right. What is your bucket size? in GPU settings? Try something like 512x512. This should make it more network friendly. The bigger the bucket the btter it distribute.

[biochemical_animations]

Thanks for the note, Dariusz. I'm still using vray 3.5701 which doesn't have a DR bucket RT CUDA mode (unless I missed it).

Is there DR bucket RT CUDA in the new builds?