Is it possble for Cycle render or other render engines to utilize
2 different GPU.
I got Geforce 1050(laptop internal) and 1070(external via thunderbolt)
in the Cycles property I was able to check both 1050 & 1070 for CUDA,
wasn’t sure if both can be used.
1050 just seems to be idle using freeware to check.
1070 also seems to only used 2GB /8GB, just wondering if there’s way to bump it up.
Quick check to see if from Raytraced point of view both GPUs are working you can click on the Cycles name in the HUD. The render device in use should be shown, for a multi-device you’d see both GPUs mentioned.
Further you can do simply timings for i.e. 1000 samples. First with only the 1050, then the 1070, then combined.
I use GTX 760 + GTX 1060 to test.
So far, the second picture appears to be lighter. Is that to be expected? I.e. that you get different results by using different hardware?
No, I accidently nudged some buttons over VNC, happens easily on the phone. Rerendering for your enjoyment.
Cool thank you.
I see both 1050 and 1070 after hitting “(Cycles)”
Hmm… The single 1070 is faster in my case,
Maybe because of the via thunderbolt external GPU setup…
Was hoping it’d be blazing fast, but it’s good to know I’d better keep just single 1070 in my case for now.
Also done 1050 on it’s own but, stopped at 1000samples. due to notable difference.
Still having problem with pop-up windows not popping up while renderview is running, and need to press Alt key to make it pop up.
Also when the pop up is open can’t seem to pause the viewport sampling.
There’s added overhead going from 1 to multiple GPUs, so yeah the results may be underwhelming unless they’re all of equal power. I had 2 970s and got a 1080ti, tried running all 3 for a while but the speedup wasn’t that impressive, I sold the 970s.
Thanks for the info.
Well, it was worth a shot and it’s good to know not to put checks on every card.
And glad I got this external GPU. I was afraid it’d be much slower going through thunderbolt.
Nice addition for any laptop users with thunderbolts.
Maybe also for mac users.
The 1050 has 640 cores. In contrast the 1070 has 1920 cores, so it is equivalent to 3 1050s.
Our current Cycles version will have fast devices idling while waiting for the slower hardware to complete their work. In upstream Cycles work is ongoing to have fast devices “steal” work from slower ones. When that work is completed and I have merged it to Raytraced you should see some gains even when adding slower hardware. Eventually better workload distribution with CPU+GPU is targetted. Future will tell.
Cool! Please keep us posted once they are ready.
