Rendering virtualization

[MajorGoob]

MODS I APOLOGIZE. I had not the foggiest clue about where to place this discussion seeing as how it deals with a billion things. It probably doesn't belong here, but idk where a better place is.

 

I had a crazy idea after reading some other forums and such while trying to figure out the best methods for making a 3d render farm. Since things like Blender Cycles (Primarily what I use. I deal with luxrender lightly but not much.) tend to perform better on many lower powered machines networked together, what if you built one super powerful dual xeon type machine, and then made a bunch of different virtual clients (2 gamers, 1 pc and 7 gamers 1 pc anyone?) to render. I feel like this might get the same type of performance, and at the same time I think its absolutely insane. Any ideas? Because 1 4000 dollar 4us are cheaper than 4 2000 dollar 4us. And quieter. And easier on the power bill.

[sazrocks]

Why not just run the render on the bare metal?

[MajorGoob]

1 hour ago, sazrocks said:

Why not just run the render on the bare metal?

If by "on bare metal" you mean just rendering with a machine without any tricks, Blender Cycles does better with a large quantity of smaller machines networked together when it comes to large rendering jobs like animation, where you have to render thousands and thousands of frames for a few minutes of video. I was thinking about virtualizing those many small machines with one big machine in an attempt to get the same effect.

[leadeater]

Unless there is a thread scaling issue it wouldn't make sense to split a physical host in to multiple VMs to render a single job. Multiple node network rendering makes sense when you need more power than a single server can deliver or is more cost effective to deploy small servers than a single larger one etc.