Instructions for setting up an extra render node?

[Booska]

I've been looking through the forum for instructions on how to go about setting up an extra machine dedicated to assisting in rendering. I'm new to all this and unfamiliar with most of the jargon, but most of the posts here are from people who seem to already know more about setting one up than I do. Can someone point me to a good source for a first timer to learn how to set one up? I'm looking for info regarding the kinds of hardware needed, what is necessary for compatibility and performance, how to get a license for the extra V-Ray node and how much it costs (I don't see those prices listed on the Chaos Group site anywhere), etc.

I'm running an i7 Sandy Bridge, Gigabyte mother board, 32 GB RAM, and a Quadro K5000 at the moment. I plan to convert it into the new render node and get a new main machine running a Titan X or 2x Titan Xs. Any recommendations on what hardware to get for the new machine and how to proceed setting up the older machine as the new render node would be much appreciated!

Like Steve Carrell's character on The Office said in one of my favorite scenes:

"I want you to explain it to me like I'm a five year old."

"..................................."

"Okay, now explain it to me like I'm a three year old."

By the way, is there a glossary of terms posted somewhere? I feel like an idiot reading all these posts with so many abbreviations and not having a clue what they mean. And I don't want to bug people by asking every time.

[Morbid Angel]

Hi,

If you read through this thread:
http://forums.chaosgroup.com/showthr...OR-SCENE/page3

there is a lot of valuable info there about various systems. Towards the end of the thread a few of us compare most recent built systems.

To answer your question about a system, first you need to decide how much you are going to spend.

Some people prefere core i7 x5690 extreme overclocked cpu. By comparison this cpu when over clocked has nearly the same output as the dual e5- xeon 2680v3, but it costs about 1/4th of the xeon cpus.

Here is some info about it:

http://ark.intel.com/products/82930/...up-to-3_50-GHz

Knowing what your budget is, you can then pick out mother board and ram. Do you want your system to be fast? if yes, don't cheap out Solid motherboard which allows for good OC and solid cooling solution which will keep that cpu stable are essential components.

Another thing to consider is SSD Hard drive. You can just google that on your own. Better performance then regular HDD.

For the second part about render node, are you going to use it in GPU mode? sounds like you are if you plan to put dual titan setup. But I would consider geforce 980's series, they pack a lot of ram and more cores then titan and if you look at gpu performance benchmark titan doesn't do so well anymore.

If you setup for DR - distributed rendering its quite simple. You install 3ds max, vray on your second machine and just launch vray spawner (from start menu/vray's install folder).

Once spawner is running, you can enter your render machines name or ip and it will see it in the list. Then you can start your render and both machines will commit to it. Weather in CPU or GPU mode this works same.

You probably will need some sorta network also. At first I didn't have a network when I just had two machines I simply connected them via cross over cable (special network cable) usually red color that just lets two machines talk to each other.

[Booska]

Thanks, Dmitry! This is just the kind of thing I was looking for! One question, though. Which is generally more preferable CPU rendering or GPU rendering?

[Morbid Angel]

It depends on your needs. I read more and more people these days try out gpu for architectural stuff, mainly interior renders. You still have to DR the cards though to get any decent time but for example 3 powerful cards can compute a very large image in 15-20 minutes, something that would be x10 longer on cpu.

With that said though and you can read more about gpus in the gpu section of the forum, the gpu is new, ever changing, unstable and many features are not yet supported. One driver version stuff works, next it doesn't. With cpu its slower, but way more stable and supports a much much larger area of things, like fur, particles etc.

[Booska]

Sorry, "DR" the cards? I do large stills with no animation. In fact, I just finished one of Lincoln at 7600 x 5400 pixels that took 4.5 days to render. Just an odd coincidence that it finished on the holiday today. I would LOVE to speed these renders up, especially since I can only have 32 GB max on my current motherboard and this scene took up 29.6 GB of it and left nothing useful to run anything else. So my computer was basically down for the last 4 days.

I suppose my needs would be similar to those doing architectural vis, since I assume their main interest is in rendering out print res images and very little animation. But I do use hair and fur, particles, sss, etc quite a lot. So if I were to get a couple of fast cards, would they not be much help in rendering that stuff?

Thanks for the info!

[Morbid Angel]

Get one card, experiment with it first before making a decision on the gpu. Gpu is limited by its ram big time. So you can't get 30 gb render into it

Sounds like you need lots of ram for your work. For the longest time I had 24 gb, but got tired of it and spent $400 on 100 gb of ram so worth that investment!

[Booska]

I would love to have 100GB of RAM on my machine, but not every board can take that much. Any suggestions on a good one? What do you think of the Asus P9X79-E WS?

When you say that a card is limited to its onboard RAM, what does that mean exactly? I thought opened files were loaded into the regular RAM. And if I had 2x GTX 980 Tis with 6GB each, would that actually mean that I would get 12GB loading capacity out of them, or does it not work that way? Is there any difference between the Asus version of this card and the EVGA, Zotac, MSI, etc versions?

Sorry, I should probably be pointing these questions at the appropriate threads...

[Morbid Angel]

Any time you get a server grade motherboard it should support a lot of ram. When you load the scene into your application, yes its loaded into your regular ram. But when you send it to gpu device it needs to process it into its own ram. So your ram in main board does not play any role there. This is a huge limitation for gpus today. For the gpu, each card has its on vRam and you can't share the ram to make more ram in total. Same thing for cpu. If you submit a DR job to two machines one has 16 gb the other 128 gb and your job takes 64 gb then guess what the machine with 16 gb won't be able to render and crash most likely.