PR: Maya 5 with Quadro FX Smashes Rendering Barrier

[CGmonkey]

OMG.. I want a Quadro FX and I want Maya 5 :drool:

Ah btw, this is my 800th post! :bounce: I need to get a life.

[GregHess]

Its not only for FX's. It works with other cards as well. Just PR :).

[elvis]

i'm very keen on seeing this rendering in action. the sheer variety of accellerators and driver quality alone is enough to make me cringe at the results that might end up coming out of this thing.

still, it is a step in the direction. the abiliity in terms of hardware has been around since the first T&L cards hit the market, so it's good to see the professional 3d market catching up.

[CgFX]

Originally posted by GregHess
Its not only for FX's. It works with other cards as well. Just PR :).

Greg,

Please be careful with the clout you wield. :-)

You have no experience with it and all you know about it is from the PR you mock. :-)

Certainly architectures like NV30 and R300, that are so radically different than those before them, might have some impact on the features available and performance seen with this solution.

I would expect the 128 bit floating point precision of the Quadro FX and 96 bit precision of the R3x0 to buy you something over earlier architectures. If only because there will need to be less pixel format conversion for doing near-final quality and HDRI work.

But, as I said, this is just the first step on the way to the point that every pixel rendered in a piece of film, video, or game content will be done on an array of GPUs and not CPUs.

[GregHess]

CgFx,

I wouldn't state it unless I already knew for a fact.

The Maya 5 Hardware renderer works with cards OTHER then the Quadro FX.

Just check the Maya listserv/hardware listserv.

Or I can give you some names at Pixar if you'd like to check directly.

[GregHess]

You can also...

Read the About Maya 5 PDF, Page 7.

http://www.aliaswavefront.com/en/products/maya/whatsnew/v5.0/pdf/newinmaya5.pdf

And I'll quote.

"Hardware rendering on the Windows and Linux operating systems supports: ATI FireGL X1, ATI FireGL Y1, Nvidia Quadro2, and Nvidia Quadro4 graphics cards. On Mac OS X the Geforce4 Ti is supported. "
-Alias|Wavefront

If anything all the Quadro FX does is add a specific shader or two...even the demo on Alias's website isn't run off an FX.

http://www.aliaswavefront.com/en/products/maya/movies/hardware_vector_renderer.shtml

Hope that satisfies you. Its interesting to note that the Quadro2Pro's are supported (old cards) but not the Quadro DCC's. Quite possibly the Geforce2's + Softquadro might work pretty well.

[CgFX]

Originally posted by GregHess
The Maya 5 Hardware renderer works with cards OTHER then the Quadro FX.


Greg,

when did I ever say it didn't?

All I said is that NV30/R300 are going to give you benefits in precision, performance, and supported features over older cards either immediately or eventually.

[GregHess]

CgFx,

I was responding to this statement before your post was edited.

Greg,

Please be careful with the clout you wield. :-)

You have no experience with it and all you know about it is from the PR you mock. :-)

[stephen2002]

when I saw the support for the Quadro2, I had to ask myself so what advantages with this kind of system are the new cards going to give you other than speed?

Anybody know?

[elvis]

CgFX, it seems to me you're just picking another shitfight. leave the personal attacks for another forum. if you know better, give us a link or some information instead of pulling apart others' posts.

ie: what greg did: links, facts, information.

[guestdude]

Does it support hardware acceleration for mental ray?

[elvis]

the hardware accelleration feature is a separate renderer to the standard one, as well as mental ray. check the maya website and it will show you the new renderer interface with specific and global options for tweaking and the like.

i suspect now that maya finally supports it you'll see similar hardware-supported renderers popping up around the place, especially given the level of sophisitication, programmability and precision in the new video cards that are appearing every day.

[CgFX]

Originally posted by GregHess
CgFx,

I was responding to this statement before your post was edited.

Greg,

Please be careful with the clout you wield. :-)

You have no experience with it and all you know about it is from the PR you mock. :-)


Hey! I didn't edit that. That gem is still there in all its glory! :-)

And it is true! :-)

All you know about it is the PR and the marketing video you watched. ;)

[elvis]

CgFX: greg made a pretty clear and concise statement. if you disagree with it (which your posts do lean towards) then please do feel obliged to explain your point of view.

if however you don't disagree, then feel free to scrap the ambiguous statements you're making which do nothing more than insinuate "i'm right and you're wrong because i say so".

you do seem to enjoy stirring the pot. agree, disagree, or don't post. pretty simple.

[CgFX]

Originally posted by elvis
CgFX:

How about you stop playing moderator and making it worse.

I was actually being lighthearted with Greg. You are just being an instigator.

[elvis]

hardly. and i'll stop playing moderator when the mods actually do something for a change.

[maelstrom]

http://www.aliaswavefront.com/en/press/maya/releases.shtml#

Maya 5 Hardware rendering is available for the Windows®, Linux® and Mac® OS X operating systems and works with most programmable graphics hardware. For supported Hardware and more details, please visit www.aliaswavefront.com.

there you have it...

Btw - a while ago Ati also said the Fire GL X1 would have the "unique" capability of Hardware rendering in Maya. Turns out it's just a new Maya feature. :love:

[GregHess]

All you know about it is the PR and the marketing video you watched. ;)

I already explained my position, and my sources.

Just saw you were joking...shouldn't do that too often, sometimes I read the forum when I'm tired.

Luckily I wasn't pissed off too :).

[CgFX]

Greg,

Glad you saw I was teasing.

Fact is, it isn't out yet so none of _us_ know-it-alls know it all yet. :wavey:

Although I have always assumed it would work with any cards my money is still on the idea that you will have more options available in the final render, higher performance, and maybe higher final precision/HDRI if you are using a DX9 level card.

There will be more GPU based renderers coming. Your pet peeve of people thinking the graphics card speeds up rendering will be history which was why I was teasing you. :-)

[GregHess]

Ya,

But I'm also sure the instant GPU rendering really starts to impact what we do....something else will come along which makes it seem utterly pathetic in comparison :).

Then of course, the instant GPU rendering can do SSS...who cares :).

[dmeyer]

Originally posted by GregHess
Ya,

But I'm also sure the instant GPU rendering really starts to impact what we do....something else will come along which makes it seem utterly pathetic in comparison :).

Then of course, the instant GPU rendering can do SSS...who cares :).

I'd like real-time GI....then I'll be happy. :love:

[CgFX]

Originally posted by GregHess
Ya,

But I'm also sure the instant GPU rendering really starts to impact what we do....something else will come along which makes it seem utterly pathetic in comparison :).


There is simply no way that a CPU can outperform a GPU in graphics rendering. GPUs are too focused on that task and now with the advent of prgrammable GPUs that can render any scene/shaders given enough passes, that gap will widen.

[GregHess]

I doubt it.

I think the instant gpu rendering comes of age, some sort of new rendering method will become available, as it has in the past, which invalidates the current gpu architectures.

Of course the gpu architectures will adapt, but remember, software and hardware always tend to follow the same path. Constant slower software for ever faster hardware :).

What good is GI on a GPU, when the CPU's can do RI? (Just used as an example)

(Real Illumination)

[CgFX]

Originally posted by GregHess
I doubt it.

I think the instant gpu rendering comes of age, some sort of new rendering method will become available, as it has in the past, which invalidates the current gpu architectures.

Of course the gpu architectures will adapt, but remember, software and hardware always tend to follow the same path. Constant slower software for ever faster hardware :).

What good is GI on a GPU, when the CPU's can do RI? (Just used as an example)

(Real Illumination)
I will take that bet! :thumbsup:

There isn't a rendering algorithm (global illumination, ray tracing, today's standard fair, etc.) that hasn't been implemented using Renderman. Renderman is a scene, shader, and illumination model infrastructure. Cg is very Renderman like in its structure. GPU based Renderman renderers are coming. This is where things are going and I am sorry you aren't a fan of that. :shrug:

Gfx rendering is all about vector math and highly paralleled processing. That is what todays DX9 class GPUs are and it is not the direction of CPUs. CPUs can't afford to be so application specific. SIMD stuff (MMX, MMX2, 3Dnow, etc.) is a bandaid on that issue but Intel/AMD can not afford to waste any more transistors on niche markets like photorealistic rendering. It just isn't going to happen.

Yes, new algorithms will come out but those too will be more efficient on programmable GPUs than CPUs. The major change is that Cg, M$ HLSL, and whatever becomes of OGL2 have enabled A|W, discreet, Avid, etc. to now be able to write GPU based renderers.

It is unstoppable at this point. :buttrock:

[GregHess]

It is unstoppable at this point.

I'd counter your bet. Hardware won't pass software until the day Microsoft is actually broken up into pieces.

Since that probably won't ever happen, can also try this.

The day Microsoft releases an OS which uses Less resources then the previous generation one. (Both in Disk Size/Ram Usage/CPU Usage)

[CgFX]

Originally posted by GregHess
[I]I'd counter your bet. Hardware won't pass software until the day Microsoft is actually broken up into pieces.

Greg,

This IS a software solution!

We aren't talking about fixed OpenGL 1.2 pipelines here. We are not talking about current or early real-time or near real-time rendering that uses hardwired GPU features. Maybe that is where you and I are disconnected.

You are thinking fixed pipeline (GeForce2) and I am thinking fully programmable (GeForce FX).

DX9-class and better GPUs allow a software solution that will be massively accelerated by the GPU in a way a CPU just can't do. A CPU can not be great at running Word or Excel and also at rendering the lastest and greatest global illumination algorithm.

The GPU may be able to maintain 60 fps with your scene or it may require 60000 passes to process thousands of lines of vertex shaders and fragment shaders, thus taking 60 minutes per frame. In either case it will be faster than a CPU.

[CgFX]

Originally posted by GregHess
[I]The day Microsoft releases an OS which uses Less resources then the previous generation one. (Both in Disk Size/Ram Usage/CPU Usage)
It was called Windows CE. :-)

[GregHess]

This IS a software solution!

The programability of GPU's is still limited by their hardware silicon. I still stand by my bet.

Bah for choosing Windows CE, I meant on a Workstation/Desktop platform :).

[raz-0]

I'll join in on greg's side of the argument. sure, GPUs have gotten some general math built in, but 1) it is not completely general. it is pixel and vertex shaders, if your algorithm steps outside of those basics, youa re back on the CPU until new architecture arrives. 2)just because a GPU is fast at some things doesn't mean it stays fast when you atart pushing it to it's limits. If the algorithm steps outside the instruction depth allowed for a pixel shader, it DOESN'T work on the GPU. However, even if it is under that magic number, efficiency falls off as depth of the program increases. It's part of the architecture. Just because it is a GPU doesn't mean it will be faster than a general purpose CPU for all algorithms.. Which is what greg is saying.

What is really at argument here is just one of those flip-flop questions in technology. And the nature of the beast is to flip and flop regularly. For example, serial vs. parallel busses or serial vs parallel processing. We twiddle back and forth between the two regularly in computing.

On the one hand, once we get fast hardware rendering, it isn't going to go away completley. However, it is going to go from cutting edge, to trailing the pack a bit, to cutting edge again. Just because of the time it takes to put out industry standards, impliment support in applications, and hardware lifecycle.

I won't even get into the concept of patents and IP screwing up the process. But right now it is M$ and gaming that are directing the path of cards. This may intersect with the cg world, but I don't expect the CG world to swing massive weight unless you have a lot of places building hardware render farms based off of a companies top tier cards.

But heck, it is all just speculation, we'll see where it goes. If A/W has some patents on it, it may never go anywhere as a single application that ships relatively low numbers compared to a successful game is not going to lead the industry around by the nose. On the other hand, if a whole industry gets on board and consumes a BUNCH of cards with a huge margin on them, they could be the SUVs of the video world.

[CgFX]

When I first bought Grand Prix Legends (best racing game ever) it used the CPU for rendering. Now all games (including GPL) are done on a GPU. Are you saying that is going to swing back??? :-)

Is anyone pushing the envelope more than Carmack? Is he going to swing back to using CPUs?

An extreme and unfair example but that is where we are headed with photo realistic rendering.

[raz-0]

Originally posted by CgFX
When I first bought Grand Prix Legends (best racing game ever) it used the CPU for rendering. Now all games (including GPL) are done on a GPU. Are you saying that is going to swing back??? :-)

Is anyone pushing the envelope more than Carmack? Is he going to swing back to using CPUs?

An extreme and unfair example but that is where we are headed with photo realistic rendering.

swing back? yes, it already has. It's game boy advance, web games and cell phone games. Did the industry swing completley? no, it seldom does. But a lot of places suddenly started spending a heck of a lot more money in the non-hardware accelerated 3d market. Some of it has even taken money away from the PC and 3d consoles.

Noticed I said you will never see it go completely away. however, we have already seen repeatedly newer technologies implimented in software first because the specialized hardware isn't there. This is in games where a lot less random fringy math makes it to being a cool new faddish feature like in CG. GPU are not general processors yet, so it is VERY conceivable that one can step outside of it's functioning parameters.

Alternately, since it is a is botha hardware and software issue, the twiddle might be consumer vs. gaming. If we finally back off on the 6 month product cycles, or consoles really do beat the crap out of the PC for a good long time, the cgmarket may be what defines a GPU rather than gaming.

As for games on the PC, hardware acceleration isn't going away anytime soon, but yes, it very well could. Or more precisely, the distinction between what is a cpu in your machine and what is a GPU in your machine may blur HEAVILY. Put it that way. Heck I could imagine something akin to a cool box shipping with N programmable gate arrays on some kind of a very fast bus, and your os reconfigures one or more, and your video driver steals one or more for it's use. If you had something like that, what is "hardware accelerated"?

heck, pong was hardware accelerated as long as you define hardware right. I don't rember breaking out the pencil to play, anyway. And that is not the twiddle I am talking about.

But seriously, we are going to lose the video card as we know it fairly soon simply because of form factor. Think about it. GPUs are headed towards outpacing CPUs in heat dissipation, yet we have a form factor that allows about a 2cm extension of the video card surface (excluidng space taken up by capacitors and other tall components) for a cooling solution. Not to mention, we are slapping ram on both sides, and can't fit active cooling on both sides properly. Of course, case and motherboard specs might change to work aorund it, so who knows.

[amygdalae]

Originally posted by GregHess
This IS a software solution!

The programability of GPU's is still limited by their hardware silicon. I still stand by my bet.

Bah for choosing Windows CE, I meant on a Workstation/Desktop platform :).

In my experience, I have to agree with Greg.
This touting of the days when your video card will replace an entire renderfarm are highly over-touted.

Everyone doing serious work in visual effects renders in a ridiculous number of layers because the size of the scenes, complexity, and layers of effects. Somehow I doubt a realtime hardware solution could ever effectively replace the power of multiple software renderers and banks of thousands of generalpurpose CPUs and terrabytes of RAM.

It's a pipe dream and a marketing ploy by nVidia to try to say that their stuff comes anywhere close to 'cinematic'

By the time hardware catches up with today's level of rendering in terms of matching PRMan 11 & mental ray 3.1 - we'll be on to bigger and better things and hardware will still be trying to catch up.

I might be vaguely interested if nvidia actually released a PRMan accelerator - as in dedicated hardware to actually assist in renderman renders - but even that would be a very daunting task and something that probably would not be worth what they would charge for it as a professional product. And there would be doubtless limits to what it could handle coming out of renderman.
Anyone remember the ART Pure3D & RenderDrives? Not worth the money and a terrible pipeline.

I dont even have much need for something like a renderman accelerator when I can use something like Irma - or any distributed tile-based interactive preview renderer.

As for facilities - why would they buy dedicated render hardware to stick in their machines when most places these days use multiple renderers, compositing packages, etc on their renderfarm.

I'll take general CPUs for the next several years.

The big brains at pixar & mental images (and splutterfish for that matter) have alot of cool things they're cooking up, and intel & AMD have plenty of horsepower on the way to render that stuff.

I would hate to be tied to dedicated hardware where I have to upgrade to a new piece of kit just to render the next version of the renderer I use. At least with software renderers you can stagger your old hardware with new and the new renderers come out.

cgFX - dont believe the hype. What nVidia & A|W are pitching is something entirely different from gpu assisted rendering - it's just rendering openGL basically with some advanced shading methods.

Haha - and it's not like they are the first to think of this - anyone remember discreet? Inferno & Flame systems (and combustion too) are just big openGL accelerators. Alot of Flame operators will actually slip in a fast GL render here and there into their final comps to save time over software rendering.

[CgFX]

Originally posted by amygdalae
Somehow I doubt a realtime hardware solution could ever effectively replace the power of multiple software renderers and banks of thousands of generalpurpose CPUs and terrabytes of RAM.


This paragraph shows you do not understand what I am talking about.

This will not be a realtime solution. This will not be scenes from Shrek or LOTR on the fly.

This will be those scenes painstakingly rendered over a period of time but on GPUs instead of CPUs. They will take 1 hour instead of 5 hours. Or they will take 30 mintues instead of 3 hours. Or they will take 1 minute instead of 45. Or they will render 1 frame per second, 5 frames per second, or 30 frames per second.

It all depends on the size of the scene, the complexity of the illumination model, and the number and complexity of the surface shaders in the scene.

Mental Ray, Pixar, Lightworks, etc. are all working on this.

You very much should be skeptical but that doesn't change the fact that a DX9 GPU is a better vector processor and FP math engine than a 32 bit, interger-focused CPU.

[amygdalae]

Originally posted by CgFX
You very much should be skeptical but that doesn't change the fact that a DX9 GPU is a better vector processor and FP math engine than a 32 bit, interger-focused CPU. [/B]

heh - this paragraph shows that you do not know what I am talking about. Since you cant spell 'integer'

[beaker]

Software rendering will always be a few years ahead of hardware renderers. People always want to the newest greatest technology in their rendering. GI today, some other new technology tomorrow. As soon as the hardware renderer starts doing full GI there will be another highly processor intensive technology that makes things even more realistic(or stylistic) that people will want. It will take another few years for the cards to catch up. Also the Cg language is still very primitive compared to renderman standard(pixar and pdi are extremely dependant on procedural textures because they don't do much texture painting so the lacking in the Cg language would be a big thing to them). There are huge gaps missing in it that need to be filled before it can even be considered for film production. Also I don't even know if the agp bus can handle the 1-2 gig per frame rib files and the hundreds of gigs of textures that many places use without choking.

[elvis]

when quantum computing arrives in 5-10 years time, entire movies will be rendered in less than a second on a machine the size of a grain of rice! then you can kiss silicon goodbye once and for all!

[FreeQ]

Originally posted by elvis


when quantum computing arrives in 5-10 years time, entire movies will be rendered in less than a second on a machine the size of a grain of rice! then you can kiss silicon goodbye once and for all!



"A second" is very long time for quantum. Maybe beta testing results had giving. ;)


BTW, all folks - what do you think when saying Hardware or Software rendering. IMHO rendering process has occur with both guys. Think of that.

I call that "optimizing to CPU or VPU"

Imagine IPR with Full AA, Ray&PhotonTrace at 2K resolution :drool:

[CgFX]

Originally posted by amygdalae
heh - this paragraph shows that you do not know what I am talking about. Since you cant spell 'integer'

Long standing rule of the internet: When someone switches to pointing out your grammer, spelling, or typos... you win. :-)

[CgFX]

Originally posted by beaker
Software rendering will always be a few years ahead of hardware renderers
Of course! And those software renderers will run on GPUs and not soley on CPUs.

Also the Cg language is still very primitive compared to renderman standard(pixar and pdi are extremely dependant on procedural textures because they don't do much texture painting so the lacking in the Cg language would be a big thing to them). There are huge gaps missing in it that need to be filled before it can even be considered for film production.

Could you expand on this with any details or links? Everything I have read about Cg shows it to be nearly identical in structure and scope to renderman with different syntax. RIB to Cg or ATI's Rendermonkey conversion tools are already coming out and showing up.

Also I don't even know if the agp bus can handle the 1-2 gig per frame rib files and the hundreds of gigs of textures that many places use without choking.

PCI Express is going to be the interface to and from the device. However, for rendering the scene it is in large part going to be the memory subsystem of the GPU that is the question. This is currently at 1 GHz, 128/256 bit wide. That may even be a benefit to the GPU, the fact that it is dual ported (PCIX + GPU mem) vs. the single FSB (and slower system mem).

[raz-0]

well all I have to say, is that when the ps2 was launching, sony was going to cram a bunch of their video chips into a machine for astronomical prices, and it was going to be able to render insane things in seconds. To the best of my knowledge you can't go out and buy one.

The r300 core was supposed to be able to be chained for massive hardware rendering. Still not here.

There's no doubt the GPU is looking to be where folks want to go, but then again, I can think of at least 5 or 6 of those things that really haven't panned out.

I think in the meantime what you will see come of it is a whole new apporach to preview and test renders based on industry standard interfaces. But once you make a GPU generally extensible, it has basically become a CPU. If it doesn't, it has the potential for wacky new ideas to not work on it. CG is full of wacky new ideas on a regular basis. Hence there will be software renderers for quite a while.

[CgFX]

Originally posted by raz-0
well all I have to say, is that when the ps2 was launching, sony was going to cram a bunch of their video chips into a machine for astronomical prices, and it was going to be able to render insane things in seconds. To the best of my knowledge you can't go out and buy one.

The PS2 does not have a programmable GPU.

In addition, the demonstrated cluster was for taking advantage of the vector processing of the CPU and psuedo-GPU only.

Beyond that fact, consoles are a horrible example because their clock rates are low (cost), their resources are low (cost), and their config is fixed for 3 to 5 years.

You buy a Quadro FX now, you will be able to run the same Cg/HLSL/OGL2 code on its follow-on next year.

[Mistyk]

Originally posted by CgFX
... Quadro FX ... its follow-on next year.
Will it likely be until next year before the next Quadro line will show up? Or was your point simply that the QuadroFX is future proof?

Thanks

[CgFX]

Originally posted by Mistyk
Will it likely be until next year before the next Quadro line will show up? Or was your point simply that the QuadroFX is future proof?
My point was about GPU's of DX9-class or better are going to run the coming wave of programmable GPU based renderers.

[Mistyk]

Thanks for the clarification CgFX.

[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.