Better Render Times?

[Matt Sugden]

I've been doing some reading about graphics cards at the moment as I'm about to build myself a new PC for Max.

 

It seems that there a lot of developments regarding the open GL hardware acceleration, but what isn't clear to me is whether or not any of these cards, or indeed whether there are any cards commercialy available, which will actually have an impact on the final rendering time of scenes.

 

I'm not talking about viewport rendering, I'm talking about the renders which are saved to files.

 

Ta

[luckytohaveher]

I'm not talking about viewport rendering, I'm talking about the renders which are saved to files. Ta

 

See:

http://www.nvidia.com/object/IO_12820.html

and

http://film.nvidia.com/page/home

 

This is the ONLY solution I know of that will use 'OpenGL' type of hardware acceleration for final rendering acceleration. Everything else is CPU dependent. I haven't used it but, base on NVIDIA's past track record you can be sure it will work well. Basically, buy a nice Quadro card and write a check.

 

Lucky

[Christopher Nichols]

Well yes... it is a common mistake to make to think that your Graphics card will give you better rendering times. CPU is the only thing that effects this... for now. But in the near future, GPU (graphics processor units) will soon be the main power to raytracing. Since they can do that way fater than GPU. The problem is that the AGP port (the only method currently that the graphics card connects to), is sort of a one way traffic: from CPU/ram to GPU to display (your monitor). When PCI express comes out, it will replace AGP, and will be 2-way, which will open the doors to GPU rendering and will be silly fast.

[Ernest Burden III]

When PCI express comes out, it will replace AGP, and will be 2-way, which will open the doors to GPU rendering and will be silly fast.

 

When will that be?

 

I have a Quadro, but its not on the supported list. Oh, well, one year old is usually too old these days. I'm coming to believe that having a yearly subscription to computers is necessary to keep up with tech advances. I don't have the money needed, or the time needed to buy the hardware and transfer the software, but larger firms perhaps do.

 

For anyone reading this thread thinking "if high-end cards don't speed up rendering why buy them?" the answer is that those cards speed up your day-to-day work, which is very important.

 

Imagine if you were writing on a word processor and the program had to keep pausing as the letters slowly appeared on the screen. Your writing flow is out the window. That is what happens with slow display response while working in 3D. Or put the other way, a fast card allows you to actually work while you are working, and not keep waiting for a display to catch up with your thinking.

[Christopher Nichols]

When will that be?

 

I have a Quadro, but its not on the supported list. Oh, well, one year old is usually too old these days. I'm coming to believe that having a yearly subscription to computers is necessary to keep up with tech advances. I don't have the money needed, or the time needed to buy the hardware and transfer the software, but larger firms perhaps do.

 

Well the type of card I am talking about is not used for display. Basically it would be similar to what the renderdrive people are doing. But it would use GPU's rather then their processor. This card would have no need to a image rasterizer, or output to VGA, and would have no need to RAM. All it would do is render. You may even be able to put 2 or 4 processors on one card. Basically exactly the same those PURE cards. I would suspect that the GPU would be a LOT faster then the processors used on the PURE cards. The other key to have PCI express since the card need to communicate with the ram and the CPU and it need to do it very fast. This will essentially be very specialized. But technically, if you have a motherboard with the new PCI express and you have new card that works on PCI express, and someone wrote the renderign engine to use it, you will be able to render with your graphics card.

 

For anyone reading this thread thinking "if high-end cards don't speed up rendering why buy them?" the answer is that those cards speed up your day-to-day work, which is very important.

 

Imagine if you were writing on a word processor and the program had to keep pausing as the letters slowly appeared on the screen. Your writing flow is out the window. That is what happens with slow display response while working in 3D. Or put the other way, a fast card allows you to actually work while you are working, and not keep waiting for a display to catch up with your thinking.

 

You are right about this, but I think that generally graphics cards are more powerful than needed for 3D work anymore. The old Quadro 900 are still more then enough power to handle most scenes well including dual monitors at 1600x1200. Most of the big GPU advances are really just making games faster. We will use the old quadro 750's at work and they are great. Keep in mind that I am using Maya that is WAY better at openGL support then MAX. I always thought I needed a better graphics card when I was using MAX, but it really come down to the fact that MAX is crappy at OpenGL (and directX aint really there yet). So buying a new graphics card simply helps the crappy openGL of MAX a little bet better.

[Ernest Burden III]

We will use the old quadro 750's at work and they are great.

 

Who you calling old? I'm using a Quadro 750. I am very used to how my models perform in OpenGL in Lightscape, which was never fully compliant with OGLv1.1 Yet the same data in the same form (polygons) shows how AWFUL FormZ is at OpenGL. Cinema4D is better, but still nowhere near Lightscape in OpenGL yet a little better in its own software mode. Why the hell would Lightscape be so much better at OpenGL display than the other, newer, programs? This really baffles me.

 

To the rendercards--I see what you mean. So you would have one of them PLUS a graphics card, it sounds like. Sounds great, as long as the rendering engines can do all their magic. nVidia mentions 'true displacement' which would be a nice feature to transfer time from modeling into rendering.

[Christopher Nichols]

haha... sorry for calling it old... these days a 750 is old since it is two generations back.

 

Anyway it is really very simple why GPU's are faster then CPU when it comes to rendering... They are designed to raytrace in the most fundamental way. They do it very very fast. They trace rays simply to find objects. We, the rendering folks, trace rays for a lot more (reflect, refract, GI, displace, SSS, etc...). Basiclaly we trace rays that bounce. But if we use the power of raytracing we can have it do all sorts of stuff... It would be like having a Brazil, or a Vray powered by your GPU. Those engines are complete raytracers, and use special raytracing accelerators to make them work as fast as possible on a CPU. But if they used a GPU instead they would be even faster... how much faster? I don't really know... but based on what I have been hearing I would have no doubt on the order of 50x faster or more. Don't think that those guys (brazil, vray, mental ray, and even renderman) are not aware of this. To them it is great since they can dump their raytracer and use the one on the GPU. The one that Nvidia is talking about is simply the first one that will do it on a GPU... it does not have the sophistication of a the other engines.

[luckytohaveher]

I just sold my NINE (9) renderdrive devices for this very reason!

 

With any kind of rendering via GPU, CPU, renderdrive, slide-rule -- whatever... There are four major corners to cover. 1.) Quality 2.) Speed 3.) Flexibility 4.) Price. This probably seems elementary but, let me explain.

 

1.) Quality: Quality can be achieved through two methods, via dedicated hardware based algorithms or complicated specific algorithms on generic logic gates. The better the algorithm the better the quality.

 

2.) Speed: With millions of calculations required for a great solution, speed is also very expensive in computing terms. In addition, every time speed improves the content creators simply increase quality. So getting a high resolution image or many frames quickly without compromising #1 is fleeting.

 

3.) Flexibility: The death-nail of dedicated hardware which benefit from #1 and #2 above. That is the killer for the renderdrive folks. They simply can not come up with a solution that doesn't take four days for RPC content. I had an architectural scene that had 80 trees. Scanline - 12 minutes. Renderdrive with no RPC trees - 1 minute 34 seconds on one device. Renderdrive with RPC trees - 12 minutes. Any questions? Adding feature like subsurface diffusion, radiosity, etc. become increasingly difficult because the mathematics or algorithms to create the solution simply doesn't exist or must be 'faked' slowly.

 

4.) Price: Creating silicon cost about $1Million/acre. Using generic CPUs, which are inherently inefficient, cost a great deal of money. The upside is that they remain fundamentally adaptable to new ways of doing things, added features, and creating new tricks. Of course, you need 40 machines to create any work of consequence in less than an eon.

 

In the end, GPUs are a great combination of all of these limitations. They have some dedicated hardware to get great quality. They are fast because of dedicated hardware. They are relatively more flexible because there is some generic logic available on chip and via the generic CPU. They have a huge leveraged cost basis on the engineering cost side and great economy of scale because they are feeding every single market out there -- not just 14 customers.

 

GPUs with CPU backup will become, with 18 months, the absolute best solution for rendering film quality images. Economics and engineering dictate it.

 

Lucky

[Christopher Nichols]

GPUs with CPU backup will become, with 18 months, the absolute best solution for rendering film quality images. Economics and engineering dictate it.

 

Lucky

 

 

18 months is a little short... It is actually happening now.

 

http://www.mentalimages.com/1_1_news/news_texte/040419.html

 

 

Accelerated graphics hardware rendering: Exploitation of available graphics hardware for high quality rendering, including support for OpenGL, NVIDIA's Cg 1.2 higher level language, phenomena and shade trees where hardware shaders are available. Automatic combination of hardware and software generated elements. Support for displacement mapping, order independent transparency, hair geometry, volumes, and high quality anti-aliasing

[Ernest Burden III]

18 months is a little short... It is actually happening now.

 

So we're back to cost. While this approach certainly is a huge cost savings vs. dedicated hardware made for a small minority of the market, I wonder how it will apply to architectural graphics. As a small firm, I do not have the workload (aka income stream) to justify a system that can render to film res and color depth in near-realtime (or even realtime). Currently, when I have animation to do, which is not all the time, I can get 2 - 4 minutes per frame full GI and raytraced per computer running frames at DVD res 24bit color. That is fine for doing a minute or two or three of animation. It would be swell if I could reduce that to a few seconds per frame (which I have done with some files, and then the hard-drive speed becomes the time bottleneck).

 

What ILM needs and what renderers need are in different galaxies of budget and return. Where's the venture capital for rendering?

 

So I'm reading all this with excitement, but also wondering just how much I will be able to take advantage of the advances in my current business.

[Christopher Nichols]

 

What ILM needs and what renderers need are in different galaxies of budget and return. Where's the venture capital for rendering?

 

So I'm reading all this with excitement, but also wondering just how much I will be able to take advantage of the advances in my current business.

 

YOUR biggest advantage is speed vs. quality. To be honest, YOU would have bigger advantage over a a place like ILM. The infustructure and the pipeline is so complex at vfx shops that it would take years and millions of dollars to retrain staff, update servers, and completly rewrite all the custom software that they have. Plus, what will they do with the renderman shader writers? It would be like have a cobalt programmers hanging around.

 

It is for this exact reason that software like Brazil and Vray, is used so much by archviz renderers and and only starting to make its way into film. The orphanage took a huge risk...

 

The biggest cost to you is mainly the software.... lets say $1000. Training time.. lets say 2 weeks (at the most). You could recoop that cost in a few weeks of work.

 

The biggest cost to a place like ILM is pipeline, the training... the cost of the software and even hardware is trivial. To them it would cost millions. The only reason that they would even consider using it, is not speed believe or not, it is if it would allow them to do something that others cannot. They would need it for a competative edge.

[Ernest Burden III]

YOUR biggest advantage is speed vs. quality.

The biggest cost to you is mainly the software.... lets say $1000. Training time.. lets say 2 weeks (at the most).

 

Well, that doesn't sound bad. At the moment I don't use anything that can run vRay, but either it will get plugged in to C4D or the stand-alone will come along. So I start without that already in place, and my GPU is 'old' now, as we observed, so I would be laying out many thousands of dollars. But that is easily recoverable. If it is a US$1000 - $5000 sort of proposition, I would re-think my comments about it being hard to bring into a one-man architectural CG business.

 

I hope that this fall we will have more of these ducks in a row, especially the vRay part.

[Christopher Nichols]

I think a lot of things will be revealed come this Siggraph. As far as GI rendering is concerned, I would say that it is going to be as big as 2001.

 

BTW... when I say YOU... I mean archviz people, not you specifically. When I say software, I mean any new software, not Vray specifically.