OTOY Forums

Hi, I'm thinking about buying octane renderer. For the past years I have been working on 3Ds max 2011 Design using Mental Ray, Photoshop, Mudbox, Vue Xtream etc.
I am currently using:
6core 3.2 gb processor, 32 gb Ram (extendable to 64 on my Board), 1x 3Gb VideoCard(extendable to 3 cards).
I would like to sell that card and buy 3x EVGA GeForce GTX 780 6GB SC which is a Cuda rady card with 3.5 Compute capability.
link to Amazon -
http://www.amazon.com/EVGA-GeForce-384- ... ce+GTX+780
link to cuda gpu list -
https://developer.nvidia.com/cuda-gpus

Questions (as I didnt find the answers in FAQ):
1.I constantly got scenes in max that require 6 - 15 gb RAM on average needed to render because of double hd image sizes or 40 000 000 polys to render etc. If I buy those 3 cards will their "power" somehow sum up to handle my scenes so that I have 18 GB GPU memory on my disposal or will I be able to handle scenes that require only less than 6GB?? Is the renderer as good as the weakest card installed?
2. Is tere a way to calculate how much memory scenes will require in OC based on image size in relation to poly counts?
3. Let's say I'll have to render something that requires 25GB RAM but I will have the whole scene done in OC (maps, shaders etc). Is there a way to make OC use CPU and RAM instead of GPU for those heavy scenes?

thanks

Hi,

1) No the VRAM does not add up, only the speed. Each GPU needs a copy of the scene to work on. There are 3 main users of VRAM in octane, Geometry, Textures and Render Target.
So you will be limited to a 6GB scene size regardless of how many 6GB 780's you have. The scene size is limited to the smallest of the cards used to render.

2) You can generally estimate the usage of geometry as about 150MB per million triangles (very rough). Usually polys are two triangles, to something like 5M polys would equate to 10Mil triangles which will be about 150MBx10 = 1.5GB. Textures are a littler harder because you can't use their on-disk size because octane will store them differently in VRAM. The render target is roughly guessable using 20MB per 1Megapixel. So a 4000x4000 render target is about 320MB, 8000x8000 is 1.3GB

3) No, octane can only use the VRAM of the GPU at the moment. This has been the design primarily because of speed, nothing will be as fast as purely using GPU memory.

Thanks
Chris.

Thanks for your reply Chris.

I would like to know a couple of other things:

1. Is the GPU's VRAM used only for redering or do you lose a certain amount of its processing power for other things like handling the monitor or whatever.
2. Can you create a decent architectural project or render an animation with only 6GB VRAM. I know there are cards with 8 gb and 12 gb but these cards are too expensive in proportion to their "power".
3.Let's say I'll start to optimize everythnig I can and use a lot of normal maps and displacement maps instead of geo, how will it influence the VRAM demand.
4. What about physically based shaders, raytracing, refraction, soft shadows, Sub Surface Scatering. How do these things influence the VRAM demand, can it be roughly calculated?
5.Can you render your work in passes just like in mental ray when you don't have enough memory, render 2 or 3 passes and then add them up together to get the finall image?
6. What in your opinion is the best way to be able to handle huge scenes with a lot of VRAM demand, is it even possible or is this renderer used only for small projects?

1. Is the GPU's VRAM used only for redering or do you lose a certain amount of its processing power for other things like handling the monitor or whatever.

Yes but you can disable single cards or set down render priority.
You can use your system fine during render but heavy 3D work is not possible, only if you disable display card in Octane.

2. Can you create a decent architectural project or render an animation with only 6GB VRAM.

Yes, example use 3 GB cards: http://render.otoy.com/forum/viewtopic.php?f=5&t=40983

3.Let's say I'll start to optimize everythnig I can and use a lot of normal maps and displacement maps instead of geo, how will it influence the VRAM demand.

Geo is not the main problem, you can use instances for example.
Image textures need much memory, resize 50% reduce RAM to 25%, so big textures in front, smaller textures in background.

4. What about physically based shaders, raytracing, refraction, soft shadows, Sub Surface Scatering. How do these things influence the VRAM demand, can it be roughly calculated?

A Q. for Otoy team.

5.Can you render your work in passes just like in mental ray when you don't have enough memory, render 2 or 3 passes and then add them up together to get the finall image?

A Q. for Otoy team.

6. What in your opinion is the best way to be able to handle huge scenes with a lot of VRAM demand, is it even possible or is this renderer used only for small projects?

Resize your image textures as much as possible, use procedual textures, use instances.
I saw huge projects in gallery, so no.

The next GTX 800 series comes with 8-16 GB it rumors, cards are coming in November it rumors.

Cheers, mib

Stratuss wrote:4. What about physically based shaders, raytracing, refraction, soft shadows, Sub Surface Scatering.

And direct/indirect caustics (PMC (seems be metropolis) kernel recommended) and more universal things.

Thx, I'll get that one 6GB GTX and give this renderer a shot.

For plug-ins, is it possible to run the entire render node on a separate system?

For example, if I have a GPU running MODO on my host system, can I run the Octane Render process entirely on a separate system with multiple GPUs? It's not clear to me if the host system must render (and have a standalone and network license) AND the second system would need a network license or if one standalone and one network license would suffice.

Is the license tied to a single host or does it just limit you to a host per license? I float between systems frequently and a single instance is preferred over a nodelocked approach.

Is network rendering available in the plug-in or only standalone mode? Assume the former, but want to confirm. While rendering with the plug-in, is the host application (MODO, Maya, etc) available or does one need to wait for the render to finish?

Can I render with two different architectures in a cluster? For example, Fermi in one system and Kepler in another?

Thanks,
Mark

MarkV wrote:For plug-ins, is it possible to run the entire render node on a separate system?

For example, if I have a GPU running MODO on my host system, can I run the Octane Render process entirely on a separate system with multiple GPUs? It's not clear to me if the host system must render (and have a standalone and network license) AND the second system would need a network license or if one standalone and one network license would suffice.

Is the license tied to a single host or does it just limit you to a host per license? I float between systems frequently and a single instance is preferred over a nodelocked approach.

Is network rendering available in the plug-in or only standalone mode? Assume the former, but want to confirm. While rendering with the plug-in, is the host application (MODO, Maya, etc) available or does one need to wait for the render to finish?

Can I render with two different architectures in a cluster? For example, Fermi in one system and Kepler in another?

Thanks,
Mark

Hi Mark,

1) You need both the master and any slaves to have a license (however slaves only require a standalone license).
2) They are node bound, but can be de-activated and re-activated to move them around. You can do this as many times as you like but only once per hour.
3) For 2.0 plugins, network rendering is available (eg: http://render.otoy.com/manuals/Modo/?page_id=977 ). The host application stays available just the same as normal (without network rendering). Think of network rendering as just adding more GPU's to your system, except they are across the network. Octane behaves the same, other than getting the remote GPU's to render as well, and integrating the results into the live viewer.
4) Yes, you can mix and match any GPU's that octane will run on, either in the same machine or in a slave. The only catch is that your limitations will be clamped to the least capable card (eg: if you have a 6GB Titan and a 1.5GB 580 rendering, then your maximum scene size will be limited to 1.5GB).

Thanks
Chris.

Chris,

Thanks! One last question - is there a limitation to mixing Windows and Linux nodes?

MarkV wrote:Chris,

Thanks! One last question - is there a limitation to mixing Windows and Linux nodes?

Nope, network rendering is OS independent. You can have a 2.0 plugin installed on windows using a slave on a linux machine.