So i’ve been getting fairly deep into blender now, doing it for 3 years as of March 3rd, 2024, and i’ve noticed my renders are taking longer and longer, getting into the 15-25 minutes of render time, i’ve optimized my renders as much as possible but they just take too long for them to be practical, and im looking to fix the render time. I’m starting to poke my head into animation and i really don’t wanna wait 1 week for a 25 second animation or something to finish.
So the big question, can I Dual GPU my PC to speed up render time?
My current Specs Are:
I9 13900K
RTX 3060 Ti
MSI Intel PRO Z690-A DDR4
48GB DDR4 3200mhz
CORSAIR - RMx Series RM850x
Is it possible to maybe Dual 2 RTX 3060 TI’s together to boost render speed or would it just cause problems?
And if not 3060’s, any other cards that would be good paired together as dual? 4070’s? 3080’s?
I use two 1070 ti’s. Rendering is close to 2X faster than one.
Note that the GPU memories will not be combined. (Some cards can use NVlink to do that. Most can’t.)
Don’t use CPU + GPU because that makes rendering slower that pure GPU.
Optix may be faster than Cuda. Others might know about that.
A trick for animation is to start multiple instances of Blender each rendering a different set of frames. That depends on how demanding your scene is to the card(s).
you can for sure squeeze more benefits from extra hardware.
I used to use two radoen duo polaris. each one have 2 gpus in it so i was rendering with 4 at the same time.
recently i was using a 4090 with a 3090. but it was not scaling well. if not the same power sometimes it’s not a good idea cause the faster one will wait for the slower one.
but extra GPU will come handy for animation. make your tests and if rendering an image doesn’t scale well with the two GPUs in parallel (not close to two times faster for example if they have the same power) you can split your animation in two separate jobs each gpu will work on half the animation, which i will always do if working on animation.
sometimes when rendering in passes i split my scene into two files where the non heavy to compute passes will run on the slower devices and the more powerful one runs the other passes.
another benefit, is if you have a render that is not in a hurry, and you need to work on the same time (different project/ next shot) you can start rendering on one while using the other one for view-port preview and stuff like that without lag
To give you another lead, you could also buy processing power on a renderfarm service. (But honestly a few minutes for a final render is completely normal. Ten years ago that wasn’t at all possible ! doesn’t invalidate your concern but just to give perspective)
Like Roken’s 3090 you may do better with one giant card than two others.
oo_1942 is right. Two cards won’t double the performance. Mine give close to 2x, about 185% or more, depending on the scene.
Consider the power requirement of giant or multiple cards.
You can run multiple instances of Blender on one card. I’ve simultaneously had 3 instances on one card and 2 on another. That was with eevee. My main concern was if GPU memory could accommodate another instance.
You could get a second 3060 Ti and it would speed things up in a number of ways, depending on how you want to configure and make use of it (both cards on a single image, two instances of Blender with one rendering using a single card while working ‘normally’ using the other card).
However, a couple of things to consider. A 3060 Ti isn’t exactly super fast to start with, for example, 2 of them are around the same as my single 3080 Ti, while a 4070 Ti is actually faster while using less power. Then of course a 4090 just blows it all out of the water, being over twice as fast as my 3080 Ti.
Then there’s the second issue, the 3060 Ti is only a 8GB card and having two or more, still limits you to 8GB, so if your renders as getting more and more complex, you are going to hit that wall.
Even 12GB in say a 4070 Ti isn’t ideal, but it is at least better, with once again, the 4090 and 24GB being the real solution.
Either way, don’t do anything just now, since in a couple weeks, Nvidia will expected to announce a ‘super’ series refresh and there’s a fair chance there could be something like a 4070 Ti Super with 16GB for around the same price as the 4070 Ti currently.
I’m running some tests for multi-instance render with Cycles and so far the results are not good. I think my system may be bottle necked since system RAM, cpu, or disk pegged at some points. So my results may not be representative of higher spec machines. Will publish results.
Two cards is nearly 2X as fast as one for single instance using both cards so that is a benefit.
Thank u all for pitching in, for the most part what i understand is that just getting a much larger card would be ideal and if im still suffering from long render times in the future that doubling up on a large card and just splitting instances would be my best bet. thetony20 also mentioned that they may be releasing Super series soon so i’ll keep a eye out for that since it won’t be just hole on my wallet to buy a 4070 Ti Super.
