Building a renderfarm based on Octane

[sherjil]

I have few questions ... I want to build a renderfarm based on Octane render .. for TV channel ... We use maya & Max ... HP workstation / and soon switch to mac Pro's ... we already have renderfarm based on Cpu's 10 racks with dual quad core Xeons...

----- 1 ---

which is better choice [ Quadro Plex ]

http://www.nvidia.com/object/product_qu ... s4_us.html

Or [ Tesla ] (i think this one)

http://www.nvidia.com/object/personal_s ... uting.html

----- 2 ----

octane rendering speed depends on what ?

Number of GPU card or something else (because i have two fx5600 quadro but if feel no difference when i put both card on render)

---- 3 ----

and will Octane in future support to distribute each frame on separate GPU card to render animation on a network

or

combine (each or multiple) GPUs from multiple computer on netwrok to give ultimate power in a user frienldy Interface (something like netwarrior http://www.randomcontrol.com/tutorials-netwarrior )

[radiance]

Hi,

These are high-end solutions, you might want to review them with your supplier, the only advice i can give is that you should go for a fermi architecture,
and the GFLOPS rating is a pretty indicator of the speed of the card for octane render.

I suggest you contact your supplier and get quotes and make some comparisons, we can't help you much with this.

Regarding your other questions, make sure you disable SLI or otherwise you won't see much difference in render speed with dual GPUs,
and you must also have a decent scene at a decent resolution otherwise you'll just near the tonemap/vsync limit and you won't see the difference on simple/fast scenes.

Regarding network rendering, this is something you do with a network queue manager instead of a render engine.
Any studio that does animation usually runs a commercial or free (like drqueue) network queue system that renders animation frames, and octane can be scripted into it.

Yours,
Radiance

[sherjil]

Hi,

Regarding network rendering, this is something you do with a network queue manager instead of a render engine.
Any studio that does animation usually runs a commercial or free (like drqueue) network queue system that renders animation frames, and octane can be scripted into it.

Yours,
Radiance

As Octane is standalone app .. I think it would be better if you have your own network rendering app ... Because I don't trust 3rd party rendering systems (like muster for maya) they are full of bugs and errors ...

if you build a distributed rendering app within Octane it will be more useful.. it can collect Graphics cards from netwrok .. and use them to speed up single frame render as well as distribute animation on each GPU

if octane was integrated into max /maya I would love to use backburner ... But Octane is standalone So my suggestion is to you is ... Please make Octance a proper Application with less headache for users who can't script or hire someone to do something specific

[radiance]

Hi,

Regarding network rendering, this is something you do with a network queue manager instead of a render engine.
Any studio that does animation usually runs a commercial or free (like drqueue) network queue system that renders animation frames, and octane can be scripted into it.

Yours,
Radiance

As Octane is standalone app .. I think it would be better if you have your own network rendering app ... Because I don't trust 3rd party rendering systems (like muster for maya) they are full of bugs and errors ...

if you build a distributed rendering app within Octane it will be more useful.. it can collect Graphics cards from netwrok .. and use them to speed up single frame render as well as distribute animation on each GPU

if octane was integrated into max /maya I would love to use backburner ... But Octane is standalone So my suggestion is to you is ... Please make Octance a proper Application with less headache for users who can't script or hire someone to do something specific

Hi,

Since you speak of expensive renderfarm systems for a TV channel, it should'nt be that complicated to setup a propper renderqueue manager not?
It would be like setting up a business network without a fileserver...

Also, we can't make a animation rendering queue as the animation data is not in octane but in the host app, therefore the host app needs to be executed to export the frame on the node in question, requiring a normal renderqueue manager.

Radiance

[sherjil]

Hi,

Since you speak of expensive renderfarm systems for a TV channel, it should'nt be that complicated to setup a propper renderqueue manager not?
It would be like setting up a business network without a fileserver...

Also, we can't make a animation rendering queue as the animation data is not in octane but in the host app, therefore the host app needs to be executed to export the frame on the node in question, requiring a normal renderqueue manager.

Radiance

I don't only speak for TV Channel systems ... Consider a situation ... I have Mac pro , plus i have Two PCs with good GPUs & 3 laptops with cuda based GPU ... it would be nice to have Octane Network manager which can collect GPU from each computer and do render .. like Netwarrior & maxwell render .. its a good idea ... just consider it .. many of artist will agree .. you can sell one master license with 5-10 slave license for home users ... and charge Companies more for each slave license .. Just an idea

[havensole]

To render a single frame using networked resources would be very difficult I imagine. Renderfarms use a cluster configuration or a grid that makes the OS think that the CPU speed is the sum of the cpu's in the network. Even this requires a lot of configuration and programming, and support on the rendering software side. This works fine as the cpu speeds and ram bus speeds work ok with network speeds, though the network interface is generally the bottleneck. When you consider the blazing fast speeds of the gpu this just becomes unrealistic at this point. If transferring data between the host ram and the gpu's ram isn't fast enough how could transferring over a network work? The best bet would be to setup a network that renders individual frames on each unit and something that compiles the frames at the end of a job. Cubix has a pretty good setup for just this type of thing. The big thing would be the host application and it's exporting of frames. Being as RIB and Collada aren't integrated yet, and how they will be integrated still up in the unkown, this is your next bottleneck. If you know someone who is decent at programming you could have someone write simple script to automate this process, depending on your host app and whatnot.

[sherjil]

To render a single frame using networked resources would be very difficult I imagine. Renderfarms use a cluster configuration or a grid that makes the OS think that the CPU speed is the sum of the cpu's in the network. Even this requires a lot of configuration and programming, and support on the rendering software side. This works fine as the cpu speeds and ram bus speeds work ok with network speeds, though the network interface is generally the bottleneck. When you consider the blazing fast speeds of the gpu this just becomes unrealistic at this point. If transferring data between the host ram and the gpu's ram isn't fast enough how could transferring over a network work? The best bet would be to setup a network that renders individual frames on each unit and something that compiles the frames at the end of a job. Cubix has a pretty good setup for just this type of thing. The big thing would be the host application and it's exporting of frames. Being as RIB and Collada aren't integrated yet, and how they will be integrated still up in the unkown, this is your next bottleneck. If you know someone who is decent at programming you could have someone write simple script to automate this process, depending on your host app and whatnot.

hmmm ... I don't know much about gpu ram & host ram speed ... Yeah the network based solution is mainly targeted to animations which really take time ... and as far as Cubix is concerned its good but not for animation purpose & future expansion ... how many Cubix you can add to a Workstation? and if someone else need to render from network ...

And for network speed being bottle neck .. i don't agree ... Now 1Gbps is normal which will translate to 128MB ... which more than enough to transfer frame from slave machine to master machine ...

im arguing just for information sake ( really want octane to be more powerful and versatile )... u might be right .. i might be right .. or v both r right at our points

[havensole]

That is true about the network speed, but you are still needing information to be processed by the host before it can go to the network. I am still pretty sure that the pcie bus speed is way faster then a 1gb network port. Something in the realm of 8gb/s if I remember right. The issue isn't entirely port speed, but bus speed inside the system. GPU's can talk to each other in the same system because the bus speed between them is so huge, but when you try to post something to the system's ram you run into a bottleneck. I agree that this would be awesome and you have my brain working now. I still think it would be better to have a script to send the individual frames out to different systems to render. This would make it more reasonable if you need to do multiple jobs where you only want to take up half the farm versus needing to configure Octane in multiple instances of the software to use specific gpus. Just seems like less of a headache from a management side.

As far as the Cubix systems go, you can have as many units as you have pcie slots. I know they make a configuration that is somewhere like 20u spaces that is a few cpu's with a few xpander units (http://www.cubixgpu.com/Files/pdf/GPU-X ... 060110.pdf).

[nicasoft]

But how does otoy apply licenses on a renderfarm? I mean if I have 1 server mono-task
and suppose 5 different users connecting to the same server, is that 1 licence or 5 licences?
thanks in advance.

[FooZe]

Hi Nicasoft,

There are a couple of parts of the terms and conditions of sale and the standalone edition license agreement that are relevant here:

Terms of Sale:
...
"Installation of the software on a client network or the like enabling more client users to simultaneously access the software than the number of workstation licenses acquired is not permitted."
...


Standalone Edition License:

YOU MAY NOT:
...
- Use the Software on a renderfarm or cloud rendering system of any kind where the use of the Software is not for private purposes, e.g., the Software cannot be made available for use to third parties by providing access to end-users to it on a commercial render-farm or cloud service.

- Sell, rent, lease, sublicense, commercially distribute, charge for use of or access to the Software, or otherwise in any manner make money directly or indirectly from copies of the Software or any use of or access to the Software.
...

Thanks
Chris.