Ooooh!

[zackepceo]

I just bought Vue Espirit. The interface is quite nice, an easy enough jump from Bryce. I am loving all the extra options, too. GPU rendering is very cool. But there's a problem. My GeForce 4 is slooooow rendering the enormous files that I love to play with in Photoshop. So, what's a good, cheap, high-quality graphics card well-suited for OpenGL?

If Jack or Sam could spare one of their old video cards.. :p

[SamuraiCatJB]

I don't think my 5600 is still back in the closet.... I already gave it away.... :(

[SamuraiCatJB]

most cards will do OpenGL now.... nVidia is still the best at OpenGL, ATI slightly better at Direct3D... but ATI is generally cheaper.

ATI Radeon X700 PRO $165
ATI Radeon 9600PRO $104
nVIDIA GeForce FX5700LE $119
nVIDIA GeForce 6600GT/500MHz $222

I own a 6800GT and wouldn't trade it, but it ain't cheap. :)

[zackepceo]

I can wait.. it's still a WHOLE lot faster than Bryce 5's pokey old raytracer.

[zackepceo]

I have a 6800GT in the other PC.. but it's not really avaliable. On a nonrelated note about the 6800GT, the two fans in it both commited suicide at the same time. This is the BFG brand fans that come stock.. how weird. I got a silent cooler for it. Now I just have to cool my PC.

[zackepceo]

Is there a difference between the Radeon 9500 and the 9550?

[zackepceo]

Here's a sad example. I just made it, the first thing. It took 22 minutes to render. Not bad considering that it would take at least 3 hours on bryce.
It looks like some surreal thing an SGI computer would display when it boots.
http://zackwp.ath.cx:81/images/vue1.jpg

[SamuraiCatJB]

the question is... is it really using the graphics card to render it or your CPU? most raytracers do not use the graphics card except to rasterize (not 3D render), they generally use your cpu.

[zackepceo]

Yeah, Bryce used the CPU (poorly). Vue is fully OpenGL rendering.

[SamuraiCatJB]

Yeah, Bryce used the CPU (poorly). Vue is fully OpenGL rendering.

I just peeked... only in Preview. Preview is OpenGL

Advanced OpenGL Engine
Previewing is done through a state-of-the-art dual resolution OpenGL-based technology that provides both quick and detailed previews of your scenes.

Brute Force
Vue 5 Esprit is designed to make the best of your hardware. It supports dual processors and is optimized for P4 HyperThreading and G5 processors.

so the actually rendering will be purely CPU, but the preview during edit is OpenGL.

[zackepceo]

That sucks. Nevermind then.

[SamuraiCatJB]

That sucks. Nevermind then.

rendering using GPU assist is relatively new, pretty rare still, and pretty much only found in 6800 series and equivalent ATI systems that have 128bit color rendering. (of course ATI is not 128bit color but rather 72bit color, but that is splitting hairs unless you are picky like me). nVidia even offers a new software package to assist in GPU based raytrace rendering, but it is expensive.

[neogin]

the two are flirting .. :p

i think nvidia is better for openGL .. heard from many reviews ..

the best cheap card .. hmm .. depending on the deal ..

http://www.marsanomalyresearch.com/evidence-reports/2004/070/moon-banding.htm

[zackepceo]

Vue was a stopgap anyway for me to get de-familiarised to the Bryce UI. I'm going to be buying 3D Studio MAX soon enough.

[Tam Hanna]

Hi,
this looks cool. Compliments! BTW, why doesn't anyone use Blender here, the free and opensource 3d renderer?
Something else-isn't CPU rendering more exact than having the GPU do it? I think that I once read about that it a german newspaper called c't...

[jjesusfreak01]

Here's a sad example. I just made it, the first thing. It took 22 minutes to render. Not bad considering that it would take at least 3 hours on bryce.
It looks like some surreal thing an SGI computer would display when it boots.
http://zackwp.ath.cx:81/images/vue1.jpg
Took about three hours to display on my computer screen. Just kidding. The foreground looks nice, but the background is really pixelated.

[zackepceo]

Hi,
this looks cool. Compliments! BTW, why doesn't anyone use Blender here, the free and opensource 3d renderer?
Something else-isn't CPU rendering more exact than having the GPU do it? I think that I once read about that it a german newspaper called c't...
Have you ever tried to use Blender? I found it a phenominal pain. GPU rendering can be made as exact as you want it to.

[SamuraiCatJB]

Have you ever tried to use Blender? I found it a phenominal pain. GPU rendering can be made as exact as you want it to.

yes, and no.... You are still limited to floating point accuracy as far as dropping bits every iteration. One of the problems with iterative and purely mathematically long techniques is that you drop or gain half a bit every mathematical operation that requires rounding. ATI is 24bit floating point process in its math, although it takes 32bit input, it converts it upon entering the processor and exiting to make you believe it is fully 32bit. nVidia's latest generation is 32bit IEEE float all the way through.... However, CPU's are double precision and soon long double. I am still hoping for the first double precision GPU, but I am not holding my breath.

For most people's work, 32bit per component (128bit 3color+alpha) is quite enough. So yes, you can make it as accurate as possible up to the bit resolution of the machine. ATI offers "fast float" of 12bits and nVidia offers "fast float" of 16bits (half-float), both are still better than 8bit per component. and you can choose what level of accuracy vs. speed you want to use. :) You just cannot exceed the 32bit IEEE single precision float accuracy.

[JackAxe]

GPU rendering is super fast, but it blows chunks compared to dedictated hardware and software rendering. At least most of the time. Some results in Maya are actually just as good as the final, but this isn't often. :)

I tried out Blender a few time, it's excellent for being free. It doesn't like my new screen though. Even the latest version is unusable now. I originally downloaded it because it supported DX export on the Mac, but it never worked right for me. I was hoping this new version would work, but it's has display issues. Bumer though, since I had to buy Deep Exploration for my PC just to do that. I'm hoping future releases will fix it.

ATI cheats with it's less then 32bit tatics :D

<]=)

[jjesusfreak01]

I wish I really knew what everyone was talking about, but this off topic, so I dont have to know everything.

[SamuraiCatJB]

I wish I really knew what everyone was talking about, but this off topic, so I dont have to know everything.

unfortunately Jack and I are in the graphics biz, on very different sides. So sometimes you have to kick us to bring us back to earth on topics that matter to us.

Raytracing is the mathematical process of tracing reflaction and refraction of the path of a photon (radiosity is a raytracing option that actually deals with distribution of energy over distance of that photon). Since you are tracing the path of light reflecting off objects, including secondary, tertiary and other reflections to reach the eye, it is an extremely time-consuming process, and very, very mathematically dependant thus traditionally done on your central processing unit of your computer.

GPU rendering is the same process sent to the graphics card. Latest generation graphics cards introduced the capability of mathematically processing a floating point number (7 decimals of precision for nVidia graphics cards), which is 4 times more than previous generation graphics cards. This means the result can be closely achieved by doing all the math on the graphics card. What no one has known is that the graphics card due to multiple internal processors handles math faster. But CPU's on your computer have about 31 decimals of precision, so are more accurate. Everytime you increase the math to add more realism, the central processing engine handles it much better, the graphics card has to round the result much earlier (thus the phrase dropping bits).

don't know if that helped, but I tried. :)

[JackAxe]

That helped me about the GPU. :D How many decimals of precision does a X800 have?

And since ATI is using only 24bits and 12 bits vs 32 and 16, wouldn't that give it an edge in performance since it will round the result earlier like you mentioned abou the decimal of precision thingy. So good for games, but bad for heavy work like Maya?

<]=)

[SamuraiCatJB]

That helped me about the GPU. :D How many decimals of precision does a X800 have?

And since ATI is using only 24bits and 12 bits vs 32 and 16, wouldn't that give it an edge in performance since it will round the result earlier like you mentioned abou the decimal of precision thingy. So good for games, but bad for heavy work like Maya?

<]=)

it does give it an edge in performance... it allowed it to nearly match the performance of the latest nVidia chips. :) and beat the last generation which when 32bit rendering was only an add-on. The ATI changes to the decimal are proprietary, not openly listed. it affects both exponent, and decimal in limitations. I dont' know how much for each. However, in ATI's favor, if you are doing all color calculations as 0.0 to 1.0 range, all in the decimal, then all the bits in IEEE reserved for exponent are wasted.

The advantage for me, is I can use it for far more than just light/color composition. :)

[JackAxe]

So this probably why ATI performs great in games, but slower then NVidia with Maya?

What is "IEEE?" :) I'm picking at your brain. ;)

<]=)

[zackepceo]

IEEE is a standards consortium that dictates all kinds of mostly useless geekiness.

[SamuraiCatJB]

well.... if you call useless the ability to store a decimal number in the same format across every platform known to man... then yes. :)

They are kinda like ANSI, but much, much geekier.

[SamuraiCatJB]

So this probably why ATI performs great in games, but slower then NVidia with Maya?

<]=)

well, that is part of the reason.... That is the reason why ATI performs great in games, but you can notice subtle color differences using Maya....

If you remember the argument I had last year with ATI.... ATI uses 24bit rather than 32bit representation... that's a 25% savings over nVidia. Yet it was only 10% faster. To me that meant that ATI was behind in technology, nVidia carried the lead. This generation, nVidia is 5% faster, but the same designs hold true, ATI is still pushing 25% savings on internal calculation to still come in behind....

Marketing is more important than technology. :) of course ATI's comment is, how many bits do you really need? in games where representation has no association with reality, you can probably get away with dropped bits. In Maya and scientific problem solving (there are people doing fluid dynamics using nVidia archetecture now), it is much more important to have more bits.

As I said, I am still waiting for double precision (64bit) decimal representation. But then I deal with VERY big numbers.... (the world)

[zackepceo]

Don't the Matrox cards have double-precision GPUs?

[Cyker]

You mean the Phaneleiaiaia thingy?

SamCat: Did you mean to say up there that Raytracing is what GPUs do? Because if you do, I got bones to pick with you matey :p :D

(What is the type of rendering GPUs do called anyway? Because I *know* it ain't raytracing ;))

[zackepceo]

Nooo. He said GPUs don't do raytracing because it's too mathy. I find raytracing to be an extraordinarily accurate waste of my time.

[JackAxe]

Raytracing is by know means a waist of time, at least not for everything. :p If you can uses shortcuts like a mapped refelection or a depth map shadow, then cool, but if you need the real deal like a monkey then RT all the way.

And remember this card. (http://www.artvps.com/). I would like to get one after I buy the RenderMan (https://renderman.pixar.com/) Plug-in, but only if that card supports it.

<]=)

[JackAxe]

well, that is part of the reason.... That is the reason why ATI performs great in games, but you can notice subtle color differences using Maya....

If you remember the argument I had last year with ATI.... ATI uses 24bit rather than 32bit representation... that's a 25% savings over nVidia. Yet it was only 10% faster. To me that meant that ATI was behind in technology, nVidia carried the lead. This generation, nVidia is 5% faster, but the same designs hold true, ATI is still pushing 25% savings on internal calculation to still come in behind....

Marketing is more important than technology. :) of course ATI's comment is, how many bits do you really need? in games where representation has no association with reality, you can probably get away with dropped bits. In Maya and scientific problem solving (there are people doing fluid dynamics using nVidia archetecture now), it is much more important to have more bits.

As I said, I am still waiting for double precision (64bit) decimal representation. But then I deal with VERY big numbers.... (the world)

ATI is just doing what they need to get that extra edge, wether it be dumbing down certain shaders or only going 24bits of the way. They always give 75%, so they can finish the job faster. :) Does that sound about right? :D

This would also be why Motion (http://www.apple.com/motion/), which requires Hefty GPU usage performs faster on NVidia.

<]=)

[SamuraiCatJB]

You mean the Phaneleiaiaia thingy?

SamCat: Did you mean to say up there that Raytracing is what GPUs do? Because if you do, I got bones to pick with you matey :p :D

(What is the type of rendering GPUs do called anyway? Because I *know* it ain't raytracing ;))

GPU's "can" do raytracing. Their default function is a simplified 3D visualization with LOTS of shortcuts for lighting and other effects. Usually you can stack up layers of textures to achieve appearance of raytracing by manually calculating lighting/shading information into textures....

However, the latest generation GPUs are floating point streaming processors as well as standard GPUs. You can "force" the GPU to do raytracing. http://www.ce.chalmers.se/edu/proj/raygpu/

Since it is a math streaming processor it is not limited to graphics at all. You could solve advanced fluid dynamics http://www.gpgpu.org/cgi-bin/blosxom.cgi/Scientific%20Computing/index.html

You can have a particle engine running completely on your GPU with acceleration/velocity and collision dection all happening without interferance or help from the CPU. http://www.gpgpu.org/cgi-bin/blosxom.cgi/Scientific%20Computing/index.html

with floating point, you are limited by your imagination, and your budget for time. ;)

[SamuraiCatJB]

Don't the Matrox cards have double-precision GPUs?

no... they, like any OpenGL 1.4 and above capable card, can take "input" in the form of double precision, but it is immeadiately truncated to single precesion by the GPU and handled as a single precision object. Matrox and nVidia do full 32bit float now, ATI still truncates to 24bit. but no one, as yet does more than allow conversion from double to single as an input process. Which means you are sending twice the data to the GPU of which it is throwing half of it away... but at least they allow receipt of double precision. That is the first step to forcing double precision, maybe by OpenGL 2.5... I am not holding my breath.

I just double checked at matrox.... nope, strictly floating precision max. they are the only ones that support 9mp displays and 10bit output. (everyone else regardless of internal calculation support has to display at 8bits due to monitor limitations... there are 16bit per component color monitors HDR moniters, but they are few and far between -- 10bit grey displays are common in medical imaging).

[SamuraiCatJB]

ATI is just doing what they need to get that extra edge, wether it be dumbing down certain shaders or only going 24bits of the way. They always give 75%, so they can finish the job faster. :) Does that sound about right? :D

This would also be why Motion (http://www.apple.com/motion/), which requires Hefty GPU usage performs faster on NVidia.

<]=)

Right, ATI is winning the input-process-output market by cutting the process to 24bit. nvidia would be 32bit I/P/O, but ATI is Input 32bit, trim, process 24bit, upgrade to 32bit, output... the longer the process the faster the ATI card is.

Motion, and other in system renderers are a:
input-process-buffer-process-buffer-process-output based system.

Now if you run that on nVidia, there is no difference between the two, its 32bit all the way through.
But if you stop to think about ATI's converting to and from 32bit at every buffer and I/O ATI's trim to 24bit becomes this:
input-trim-process-upgrade-buffer-trim-process-upgrade-buffer-trim-process-upgrade-output. Suddenly ATI's timesaving trim to 24bits is holding back its speed every time you buffer the result.

[JackAxe]

Arrr, they be shooting them selves in foot. :) I'm pretty sure Tiger Core will also perform much better on NVidia, since Apple's had more time to work with their cards and ATI is slower for Quartz redraw.

<]=)

[zackepceo]

Raytracing is by know means a waist of time, at least not for everything. :p If you can uses shortcuts like a mapped refelection or a depth map shadow, then cool, but if you need the real deal like a monkey then RT all the way.

And remember this card. (http://www.artvps.com/). I would like to get one after I buy the RenderMan (https://renderman.pixar.com/) Plug-in, but only if that card supports it.

<]=)
That's a reealy nice card. It has to be if it's that big..

[JackAxe]

Yep, you're paying for size alone. :D

Freak, my TI 4600 and the 6800 U are also full length cards. It's fine for the G5, but with my G4 it created a wall that made setting up my extra ATA133 drives a nightmare.

<]=)

[SamuraiCatJB]

Arrr, they be shooting them selves in foot. :) I'm pretty sure Tiger Core will also perform much better on NVidia, since Apple's had more time to work with their cards and ATI is slower for Quartz redraw.

<]=)

like with anything, somethings it works, somethings it doesn't. :)
For games if you design the software right, you can make it work for you. I don't write games, so it doesn't work for me. :)

[SamuraiCatJB]

Yep, you're paying for size alone. :D

Freak, my TI 4600 and the 6800 U are also full length cards. It's fine for the G5, but with my G4 it created a wall that made setting up my extra ATA133 drives a nightmare.

<]=)

bigger really is better... :) http://www.extremefunnypictures.com/funnypic700.htm

just ask about. :p

[zackepceo]

The 6800 DDL is supposedly a huge card.. the 6800GT isn't much bigger than a standard card, and I just ordered a DDL 6600. I wonder how big it will beeeee...

[JackAxe]

bigger really is better... :) http://www.extremefunnypictures.com/funnypic700.htm

just ask about. :p

*LOL* :D

<]=)

[JackAxe]

The 6800 DDL is supposedly a huge card.. the 6800GT isn't much bigger than a standard card, and I just ordered a DDL 6600. I wonder how big it will beeeee...

The 6600 is the same size the 6800GT. :) My friend has a 6800 GT in his Shuttle case. The GPU is exactly the same as the 6800U and FX4000. I bet you the 6600 is also the same, but I don't know that for a fact.

Thuis is how big my DDL and TI4600 are. (http://www.chinadaily.com.cn/english/doc/2004-05/14/content_330829.htm)

<]=)

[zackepceo]

Even the DDL 6600?

:p
I'd really like to see the Pyramids of Giza from space.

[JackAxe]

That pyramid is about the size of the heatsink on the 6800U, maybe a little bit smaller. :)

See, it's normal size, none of that full length wall stuff.
http://www.beyond3d.com/previews/nvidia/nv43agp/index.php?p=2

<]=)

[zackepceo]

Aww, I wanted a behemoth card. I guess I could get a Voodoo 5 from eBay... :D

[SamuraiCatJB]

Aww, I wanted a behemoth card. I guess I could get a Voodoo 5 from eBay... :D

just go for thicker.... with a 4400G you will feel like it has a full length card but cut in half and stacked like a sandwich. And of course your pocket book will be much lighter, so you will feel very important for buying the most expensive item available. ;) http://www.nvidia.com/page/quadrofx_3000g.html

;)

[zackepceo]

Eek. No need for Genlock!

[zackepceo]

I know, I'll buy a PCI Voodoo 5 and put it in my Mini-ITX, resulting in the video card being significatly bigger than the motherboard.

[SamuraiCatJB]

I know, I'll buy a PCI Voodoo 5 and put it in my Mini-ITX, resulting in the video card being significatly bigger than the motherboard.

mod the case.... ;)

[zackepceo]

That would make me have to have a giant thing coming out of the top and front.
:p

I'll gladly take your Ti4600 for when I get 3D Studio Max.

[SamuraiCatJB]

your micro PC case will just look "very" excited....

[zackepceo]

I think this could be what you want.. http://www.sgi.com/products/workstations/tezro/graphics.html

[SamuraiCatJB]

I think this could be what you want.. http://www.sgi.com/products/workstations/tezro/graphics.html

puaaaalease.... I ran SGI for waaaay too long...
http://www.zianet.com/jjustinb/emoticons/spew.gif

[JackAxe]

SGI has that old IBM mainframe feel now.

<]=)

[SamuraiCatJB]

SGI has some nice new machines.... just cost per performance ratio is extremely high for what little you get....

I'd rather have the Quadro FX 4400 with 512mb, it is bloody expensive, but at least testing the 512mb board would be bloody FUN! :D

[zackepceo]

Wouldn't you need a dual-AGP/16x PCIe motherboard to use it though?

[zackepceo]

SGI has that old IBM mainframe feel now.

<]=)
Yeah.. SGI has a freaky little niche of people who are fanatical about their good-looking (but incredibly heavy) computers. Sounds similar to the Colour Classic Cult.. that I am guilty of being a member of.. :o

[JackAxe]

I really like the old SGI designs, but now they're kind of fugly, kind o like Tupaware.

The Color Classic was colorful and had lots of colors.

<]=)

[zackepceo]

http://www.stuartbell.dsl.pipex.com/PowerCC/ccphoto.jpeg

[zackepceo]

Here you can see my Mini-ITX. Not only will no PCI card fit at all, but the Voodoo 5 will DEFINATELY not fit.
http://img25.exs.cx/img25/8522/mobo5jh.jpg

[JackAxe]

*Sexy* :p

<]=)

[zackepceo]

Oh yeah.
http://img86.exs.cx/img86/7334/top7bk.jpg

[zackepceo]

I threw all the insides that originally came with the Classic away because they were MicroSPARC poo.

[SamuraiCatJB]

Wouldn't you need a dual-AGP/16x PCIe motherboard to use it though?

only for SLI cards. :)

[zackepceo]

Oh, I see. It's only the width of 2 cards, not the slottage.

[zackepceo]

If you didn't use DirectHex, couldn't you feel the GPU higher than 32-bit stuff?

[SamuraiCatJB]

Oh, I see. It's only the width of 2 cards, not the slottage.

yup, just a fat-card. :)

[SamuraiCatJB]

If you didn't use DirectHex, couldn't you feel the GPU higher than 32-bit stuff?

still dependant on hardware.... OpenGL allows "input" of double precision, but no hardware actually uses it, they all convert down on the fly. The problem is that the market is soooooo speed conscious that double precision would cut performance in half, so they won't do it.... dual support like now with 32/16/8 slows down partially to offer multi-pathing... this was why nVidia lost performance for a generation and ATI fell behind again on the catch up. adding a 64bit or 128bit path on top of the 32/16/8 multi-path would slow it down even more...

[zackepceo]

Ugh. I have to render this crystally thing in Bryce 5 because the format it uses is proprietary.. Slower than absolute zero.

[JackAxe]

That is why i have 3 PCs in here along with my Macs, for rendering. :)

<]=)

[SamuraiCatJB]

here is an interesting development.... As you may know nVidia's processor design is CISC, they have many more microcode operation commands on their GPU processor, it looks like a new microcode instruction is "coming" or already here.... I got this from the latest release for the CG compiler from nVidia and the VP40 FP40 specs.... "New ‘-fastprecision’ option for arbfp*, fp30, and fp40 profiles, to use reduced
precision storage (fp16) when appropriate. When used with the ‘arbfp*’ profile, this
option will generate the ‘precision_hint_fastest’ option in the output stream."

Note that it says that this is put in the output stream, which means the chip could decide on the fly what level of precision would produce a benefit without loss of precision.... Although an interesting "auto" shortcut from 32bit float to 16bit float, it may be the predicessor to getting me 64bit. :D After all if the decision making structure can get in the microcode without upsetting the stream, then they could introduce newer modes down the line without upsetting existing data paths. :D

[JackAxe]

CISC eeeeeeeeewwwwhhhh. How archaic. I'm a RISC fan. :)

<]=)

[zackepceo]

I like dynamic vector processors. :p

[JackAxe]

It's vectorific. :o

<]=)

[zackepceo]

I still want a Cray 2. :D

[SamuraiCatJB]

I still want a Cray 2. :D

nawwww, I'll take the stony brook university computer....
http://www.cs.sunysb.edu/~vislab/projects/gpgpu/

[zackepceo]

Not for the performace, they look soooo coooool. And what other excuse are you going to use to buy Flourinert?

[SamuraiCatJB]

hey, I am a geek why would I ever get something for it's appearance. :) ;)