Jump to content

Blender-Animations - which hardware?

cisto1999

Hi,
i want to build myself a workstation for Blender. I am doing some 3D animations - about 5-8min long (only very seldom a still-picture rendering).

I have tried to get a specific answer in special Blender-forums - but the answers are opposite - one says this the other one says that... maybe you can help me out in this forum:

Now my question: Is it better for my needs (animation) to invest in one giant powerful workstation with multiple GPUs (max. 6 GPUs) or to build 3 render-nodes with each 2 GPUs and connect them via network?
I have some concerns for each variant: The one huge WS will easily become very expensive and i have to spend all money at once. Then I have also read that mulitple render machines will scale better than one powerful.
But for the render-nodes i have the concern, that it seems to be sketchy to set it up especially working with GPUs (with different amount of VRAM and so on)... i don´t even know if that is possible at all? Does there exist a GPU based network-render engine for Blender? The pros for the nodes would be: i could expand and invest with my needs and my budget and i find them easier to upgrade (you can do one by one when new GPUs hit the market).

I would really appreciate any thoughts, recommendations and help for this since i am a noob in this field...

 

Here the answers i got so far in other forums:

 


Cycles scale well up to 3-4 cards on complex scenes and I know from Octane, Cuda scale nearly 100% up to 8 or more cards.
Cycles is not so good but the developer work on it.
One big system is much more effective than 3-4 small ones.

 


It depends what you are rendering. For previews and single frames, multiple cards in the same machine are superior. You cannot currently net-render single frames. You can have multiple machines work on them, but you do it by having them each render the same image with a different render seed, them mix them all. (500 samples at seed 0 + 500s samples at seed 1 = 1,000 samples.) You could also split an image up by tiles, but that's not really any better.

It also matters if your cards are working together or separate. Two cards working on the same render only gives you ~50-60% of the speed of the second card. For this reason, it can be better to have the cards work separately. In the case of animation, instead of having 2 cards work on the same frame and suffer a hit, it will often be better to have them work separately on 2 different frames, so each gets full power. So, maximum speed comes from an environment where you can run multiple instances of blender, with each card working on a separate frame range. That also means you can assign appropriate parts of a scene to stronger or weaker cards (such as if you had some lower VRAM cards, you could give them a renderlayer with less going on.)
Sounds like multiple nodes are the better option.

 

 

So my specific questions are: is a single WS performance-wise better than multiple render nodes if i am rendering animations (5-8min long)? And: Are GPU-based Render-nodes even possible in Blender? If yes: Which renderer i have to use?

I don´t need an advice how to build a powerful WS and so on... (i write this, because often people send me just a hardware-list instead of trying to answer my questions)

Link to comment
Share on other sites

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×