[ASyme]

Hi all-

OK, second post in as many minutes. I've got a little shop with a couple dozen either multiprocessor or i7 computers. We use Cinema 4D for our base pipeline. Vray has been a staple since it came out for C4D. Love the Vray. However, I've been smitten by Maxwell lately and LOVE the look. Planning on big thing with this rendering champ.

My new question is about net rendering. Normally with Vray if you add a dozen clients to a job it speeds the project up by 10 times or so. Great. After messing around with Maxwell, it doesn't seem to work like that. I'm getting that each computer on the net works out it's own "version" of the scene (thinking single frame here) and them combines them into a "better" hence faster version of the still. Is this correct?

So by using multiple machines it gets to a higher sample level by combining however many machines are working on it. This is hard for my head to get around. All these machines still need to do like, what 12-16 sample levels? That could still take many hours for a complex still. Right?

Am I missing something here? How does adding many machines lower the rendering time of a still?

Sorry for being dumb about this, I'd love a clear explanation. I hope I'm wrong and just don't get it yet.

Thanks!!
Alec

[simmsimaging]

It scales in a pretty linear fashion, much like Vray. There is a certain amount of overhead (transferring files/textures and final MXI's, plus merging) but the bulk of the time is really rendering in most cases where co-op makes sense, so having more boxes active will definitely pay off.

Each machine will still need to render to a high enough sample level to add up to the target res, but how far they go depends on how many machines are involved. If you have a target of, say, 16SL and 4-6 nodes, then each will probably need to get to around 11-13, maybe a bit more or less, to get a co-op result of 16. However if you have 20 nodes they will each need to get to less (which is why renders are so damn fast at places like Rebus)

Bear in mind that the SL time is (I believe) more or less exponential, so even cutting off the render per node at 13 versus 16 is likely to be a huge time saver. For example it might take 1 hour to go from 11Sl to 12SL, but to go from 12SL to 13SL is probably 2hours, then 4 hours etc.

Also keep in mind that the more machines you throw at it the faster it will go, but the payoff tapers in the same linear kind of fashion. Going from 1 computer to 2 will halve your render time. Going from 4-8 will also halve it, but you need to keep doubling the power to halve the time, so going from 20 to 25 is far less noticeable than from 1-6 would be, etc.

/b

[Mihai]

To keep it simple, if you have 2 machines with the same hardware, you'll render in half the time compared to rendering on one machine. Maxwell scales very linearly and we've made tests to show this. There was a graph somewhere but I can't find it right now. I would say it scales better than typical bucket renderers. I don't think you'll see a falloff in the performance even if you add 20-30 machines.

You have three ways to do network rendering:
Cooperative - all machines work on the same MXS file and then merge the result of each one to give you a final MXI file.
Distributed - every computer works on its own MXS file, for example if you have several camera views in the same scene.
Animation - much like distributed with the difference that you can specify a range of frames (effectively a range of MXS files) and each machine renders its own MXS file.

[simmsimaging]

To keep it simple, if you have 2 machines with the same hardware, you'll render in half the time compared to rendering on one machine. Maxwell scales very linearly and we've made tests to show this. There was a graph somewhere but I can't find it right now. I would say it scales better than typical bucket renderers. I don't think you'll see a falloff in the performance even if you add 20-30 machines.

I didn't mean that there is an actual performance falloff, I just wanted to highlight the point that the more machines you have, the more you need to add to *see* the same kind of appreciable changes in render time. Your time savings/value *per node* begins to go down as your overall numbers increase. I'm sure that is pretty obvious to many, but I had to really think about that (not much of a math guy I guess )

/b

[ASyme]

Thanks guys for the help. It's starting to make sense. So simply, I have a still I know looks great at SL 21, but it takes 20 hours on one machine. If I net render this on ten machines it should look just as good in 2 hours? So they'll hit a SL of 14 or so in that 2 hours and they are all combined into one clean image?

Is this the thinking here?

Thanks again.

Alec

PS- Brett, that's some fantastic bottles and fluids my friend. Beautiful work man!

[simmsimaging]

Thanks guys for the help. It's starting to make sense. So simply, I have a still I know looks great at SL 21, but it takes 20 hours on one machine. If I net render this on ten machines it should look just as good in 2 hours? So they'll hit a SL of 14 or so in that 2 hours and they are all combined into one clean image?

Is this the thinking here?

Yep, probably a little less than 2 hours.

PS- Brett, that's some fantastic bottles and fluids my friend. Beautiful work man!

Thanks very much!