Nvidia Tesla C1060 for GPU rendering...?

[olson]

Since Octane Render uses CUDA, anything that supports CUDA should work including the Tesla cards. Not sure if the video memory has any impact on Octane Render performance or not. Cheers!

[tecton3d]

Since Octane Render uses CUDA, anything that supports CUDA should work including the Tesla cards. Not sure if the video memory has any impact on Octane Render performance or not. Cheers!

true!

any CUDA card will work with Octane, but it's the issue of a gaming card having enough Vram to hold a heavy (poly, textures, etc) arch-viz scene since the entire scene is held on Vram and not system RAM. In which case, the V-RAM memory issue is the same as if the renderer were software based (MR, V-ray, etc. instead of GPU like Octane) and using system RAM. So it would limit one greatly if you were working with large scenes.

prost!

[meleseDESIGN]

The Quadro FX 5800 has 4GB Ram as well and will also give you two DVI outputs.
Don´t know the price differences between the mentioned Tesla vs. Quadro FX 5800, but the Quadro makes much more sense to me. I guess you need at least a Quadro to be able to use a Tesla, probably GeForce cards aren´t compatible with Tesla expanding cards.

Try to contact the nvidia support for answers.
Maybe you get better help as well if asking in the Octane forum for more details.
They should be interested in this field as well. ;).

[tecton3d]

The Quadro FX 5800 has 4GB Ram as well and will also give you two DVI outputs.
Don´t know the price differences between the mentioned Tesla vs. Quadro FX 5800, but the Quadro makes much more sense to me. I guess you need at least a Quadro to be able to use a Tesla, probably GeForce cards aren´t compatible with Tesla expanding cards.

Try to contact the nvidia support for answers.

the fx5800 is much more expensive (+$1700 more!) than a tesla c1060 so that's justification enough in itself. I'll get ahold of Nvidia for a definitive answer regarding whether Tesla needs a quadro.

Maybe you get better help as well if asking in the Octane forum for more details.
They should be interested in this field as well. ;)

see the 2nd sentence of my first post ;)

[meleseDESIGN]

It would be nice if a GPU Renderer could make use of the System RAM and not only of the RAM on the GPU Board. Could someone please built a connection between GPU RAM and System RAM, so a GPU Renderer would be independent of the GPU RAM?!

With the Quadro FX 5800 we can see it´s not that difficult to put a lot more RAM on the GPU Board, so there are possibilitys out there allready - expansive but there. Let us see what the future brings, i´m a optimist! ;)

[olson]

It would be nice if a GPU Renderer could make use of the System RAM and not only of the RAM on the GPU Board. Could someone please built a connection between GPU RAM and System RAM, so a GPU Renderer would be independent of the GPU RAM?!

With the Quadro FX 5800 we can see it´s not that difficult to put a lot more RAM on the GPU Board, so there are possibilitys out there allready - expansive but there. Let us see what the future brings, i´m a optimist! ;)

OpenCL and I would assume CUDA as well can use both main system memory and device specific memory (in this case video memory). I say device specific memory because using the graphics card for processing is only one potential use for OpenCL. More information on OpenCL and the memory model it employs.

http://ati.amd.com/technology/streamcomputing/intro_opencl.html

[Srek]

It would be nice if a GPU Renderer could make use of the System RAM and not only of the RAM on the GPU Board. Could someone please built a connection between GPU RAM and System RAM, so a GPU Renderer would be independent of the GPU RAM?!

With the Quadro FX 5800 we can see it´s not that difficult to put a lot more RAM on the GPU Board, so there are possibilitys out there allready - expansive but there. Let us see what the future brings, i´m a optimist! ;)
The connection is already there and it's name is PCIe 16 2x
Sadly it's still far slower than a direct GPU<>memory connection and therefore not realy useable without decreasing performance seriously.
A direct connection between GPU and the RAM on the mainboard is not doable without integrating both in one device, effectively ending up with an expensive graphics card with more RAM, something we already have.

Cheers
Björn

[Jettatore]

With the prices of the consumer end cards so affordable, and also they hold their resale value, why not just try one of them out first and ensure that it is somehow not enough for your purposes?

Also have you considered compositing? There is no great need to have every single high res detail loaded into the same scene for working purposes nor for rendering purposes. If you simply use proxy stand-in's, measure everything out and then composite the image together in pieces you will save yourself a ton of head-aches and have no issues with working or rendering on large scenes.

[InfernalDarkness]

Also have you considered compositing? There is no great need to have every single high res detail loaded into the same scene for working purposes nor for rendering purposes. If you simply use proxy stand-in's, measure everything out and then composite the image together in pieces you will save yourself a ton of head-aches and have no issues with working or rendering on large scenes.

Or just use 64-bit applications. Although even so, some level of compositing will always be preferred! The post-control is much more powerful than pre-render control, for certain!

[tecton3d]

With the prices of the consumer end cards so affordable, and also they hold their resale value, why not just try one of them out first and ensure that it is somehow not enough for your purposes?

Also have you considered compositing? There is no great need to have every single high res detail loaded into the same scene for working purposes nor for rendering purposes. If you simply use proxy stand-in's, measure everything out and then composite the image together in pieces you will save yourself a ton of head-aches and have no issues with working or rendering on large scenes.
very true, I would prefer to not have to hassle with two cards as my initial interest is in tesla and it's 4gb of VRam ;) I'd like to think the new Nvidia 4xx cards will have Vram comparable to the tesla (the GTX 480 is rumored to have up to 3gb) but we'll have to wait and see.

Sure, making a model efficient is always priority but with 64 bit, it's much easier to have all the detail in the model, and since I'd be using the GPU of discussion for real-time rendering, compositing and proxy options may be more limited than with traditional softwares. As of now, I don't think Octane has a proxy function.

prost.

[DanielWray]

GTX 480 has 6GB of RAM.

I've tried Octane and it's remarkably good at memory management, I would think a 2GB card would allow you to get quite large scenes with big texture maps.

[tecton3d]

GTX 480 has 6GB of RAM.

I've tried Octane and it's remarkably good at memory management, I would think a 2GB card would allow you to get quite large scenes with big texture maps.

...all i can find is that it's being pre-ordered (http://www.sabrepc.com/p-174-nvidia-geforce-gtx-480-2gb-gddr5-pci-express-x16.aspx) @ 2gb...

thanks.

[armoos]

tecton3d!
actually I'm doing some research on using these stuff in GPU rendering at my university right now! I Have 4 Tesla c1060 with a Quadro fx 580 and I Have Demo Version of octane Renderer! However I realy need some hint on how to render sth (for example some model from maya) in Octane Renderer! if you could give me some advice I can report the result here!
I'll Appreciate any help :)

[DanielWray]

There are a few tutorials on the Octane website.

Basically you just need to import your OBJ file and click the macro icon in the node editor for a quick set-up/ render.

Then click on the nodes to change their respective settings, for example clicking on the object (from memory here, so may be wrong) should bring up the material options.

You can add various items, such as HDRI probes, image textures and various other things just by adding nodes.

[armoos]

Thanx!

but there are some issue with the work station! it's unstable and stuck in blue screens errors for practically no reason! I reinstall the windows few times and still the blue screen thing happen after a while, last time it give blue screen and restart when I try to Shutdown. I try to remove applications that may cause the problem and still got the blue screens ... I can not work with the octane renderer yet! but I'm working on it! wish me luck!

[absolute3d]

GTX 480 has 6GB of RAM.

I've tried Octane and it's remarkably good at memory management, I would think a 2GB card would allow you to get quite large scenes with big texture maps.

you can't use a gtx480 on a 24/7/365 render node...that is why you'll need a professional card like tesla/quadro ;)

you can use it for personal use, but it's hard to use it on production, because gaming card lose 15/20% performances on first year use.

[cgbeige]

if you are a licensed Octane customer (50 euro sale on now), you can get access to the plug-in builds. There is a Maya exporter that sets up multiple GPUs for you when you render to Octane. I have a basic tutorial here:

http://www.vimeo.com/12155161

Just put "0 1" in the multiple GPUs and it works (I've tested on a dual GTX 285 Mac Pro). Otherwise, the way it works for multiple GPUs within Octane is to load your mesh but don't select it, then go into the Device Manager and add the other GPUs and hten click on your mesh and it will use all:

http://grab.by/grabs/db74a956892398f00c44a60bc4cfb669.png

armos - I suggest you buy Octane and go to the real Octane licensed user forum if you want good support. The guys are very good with support there. You'll also get access to newer builds than what is available with the demo.

[DanielWray]

you can't use a gtx480 on a 24/7/365 render node...that is why you'll need a professional card like tesla/quadro ;)

you can use it for personal use, but it's hard to use it on production, because gaming card lose 15/20% performances on first year use.

I'm not sure where you get those numbers from, but I'm a little dubious to believe that.

I'm pretty certain if I had a GTX480 with reliable cooling that I could run it for production use, after all it's the same silicon in most cases (Quadro and Geforce series)

[olson]

you can't use a gtx480 on a 24/7/365 render node...that is why you'll need a professional card like tesla/quadro ;)

you can use it for personal use, but it's hard to use it on production, because gaming card lose 15/20% performances on first year use.

What? None of that is true. :curious:

[DanielWray]

Just thought I'd post this;

http://images.dailytech.com/nimage/14881_large_Dual_GPU_Fermi.png

A dual Fermi based board... It only draws around 480W :argh:

Anyway, I'd imagine that two of these boards in a well cooled system would be immense when coupled with Octane!

[vlad]

you can't use a gtx480 on a 24/7/365 render node...that is why you'll need a professional card like tesla/quadro ;)

you can use it for personal use, but it's hard to use it on production, because gaming card lose 15/20% performances on first year use.

Sources? :hmm:

[kcpr]

Hi Guys!

GTX 480 has 6GB of RAM.

I've tried Octane and it's remarkably good at memory management, I would think a 2GB card would
allow you to get quite large scenes with big texture maps.

6GB?? I've never met that card with 6GB onboard!

Could you give me a link to such model? I would buy it immediatelly ;]


Second question:

I've got GTX460 2GB and it performs well, but -YES- 2GB isn't enough for complex 3d scenes, so I have to buy at least 4GB card.

The question is: Will Tesla C1060 work togeher with my GTX460??
I have Quadro FX5800 as an alternative, but it is waaaay more expensive.

So if it's possible I would stay with Tesla C1060....

[imashination]

Theres no 6 gig gfx cards available as best I know, consumer, workstation or otherwise

[MarkusB]

Theres no 6 gig gfx cards available as best I know, consumer, workstation or otherwise

Quadro 6000, 6 GB GDDR5, price around Euro 3200,- plus tax

That are some euros to pay !

[kcpr]

OK, and how about GTX460 working together with Tesla C1060?

Will they work together?

[olson]

OK, and how about GTX460 working together with Tesla C1060?

Will they work together?

They should function together for CUDA and/or OpenCL, but be sure you have sufficient cooling and power. Multiple GTX cards might be a more cost effective option? Also don't hijack threads. ;)

[kcpr]

Multiple GTX cards might be a more cost effective option?

Yes, but until they have no more than 2GB of memory onboar they'r useless for bigger 3d scenes.

With 2GB forget about many proxies, trees, details... ;]

That's why I have to buy proffesional 4GB card.

[CGTalk Moderation]

This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.