Shader Model 3.0 Done Right? X1000 lacks "Vertex Texture Fetching."

[Tim]

Shader Model 3.0 Done Right?
ATI has been selling the "Shader Model 3.0 Done Right" slogan along with the launch of their Radeon X1000 line of video cards. (Which we still have not seen for sale without delayed delivery.) ATI proclaims that they have done Shader Model 3.0 "right" by improving dynamic branching performance and turning batches of pixels into threads. However, with some recent news that has come to light we wonder if this slogan is really correct.


According to The Tech Report, the Radeon X1000 series lack a Vertex Shader 3.0 feature called "Vertex Texture Fetching." Vertex Texture Fetching is useful when you need a vertex shader to read from texture memory. There are certain 3D effects which can benefit from this ability such as true dynamic displacement mapping. Now, this 3D effect itself may not be currently used in current games, and in fact may not even be used until the next generation of DX10 video cards. However, the case here is that ATI is claiming they have done Shader Model 3.0 "right," yet they are missing an official Shader Model 3.0 feature. This all seems a bit misleading. How can you claim to do Shader Model 3.0 right and then leave out a feature of the specification? The GeForce 6 and 7 series from NVIDIA supports this feature. So where does that leave ATI? Have they really done Shader Model 3.0 right or not?

It does appear that a workaround is possible. However this does put more work on the shoulders of the game content developer instead of exploiting an easy Shader Model 3.0 feature in their code. Isn’t the point of standards and Shader Model 3.0 to allow game content developers and programmers a standard API among graphics cards to make their job a little easier?

This may not be such a big deal if this particular 3D feature is never exploited until the next generation of video cards. But that isn’t the point. The point here is that ATI has claimed to do Shader Model 3.0 "right," and from what we have seen today, maybe it isn't so "right" at all. “Shader Model 3 The ATI Way?”

Radeon X1000 series lacks vertex texture fetch
by Scott Wasson - 12:06 pm, October 6, 2005

This question came up in the late stages of writing my Radeon X1000 series review, and I just got confirmation from ATI yesterday. Turns out that the vertex shaders in the Radeon X1000 series GPUs don't support a notable Shader Model 3.0 feature: vertex texture fetch. As it sounds, this capability allows the vertex shaders to read from texture memory, which is important because texture memory is sometimes treated as general storage in programmable GPUs. Vertex texture fetch is useful for techniques like displacement mapping, where the vertex and pixel shaders need to share data with one another.

I asked ATI's David Nalasco about this issue, and he suggested a possible workaround for this limitation:

No, vertex texture fetch is not supported. However, since the X1000 family does all pixel shader calculations with FP32 precision, just like the vertex shader, it is possible to get the same results using the render to vertex buffer capability. Basically, you do a quick pre-pass where you render to a special linear buffer in which each pixel represents a vertex. Textures can be used to modify each vertex through the pixel shader, and the result is then read back into the vertex shader. The result is fast vertex texturing with full filtering support, without requiring any special hardware in the vertex shader engine.

Note that render to vertex buffer is possible in R4xx as well, but is limited to FP24 which could cause precision issues in some cases.

Such a workaround would likely involve a performance penalty, but I doubt it would be a major hit. The larger issue is probably just the fact that the workaround would require special consideration from developers, because the GPUs lack a straightforward vertex texture fetch capability.

I'd say that's a pretty big letdown for me personnally....

I'll put the sources up in a sec....

[H]ardOCP
Techreport

So why is ATI always NOT telling these things? Instead of doing it right they just branded it SM3 the ATI way....

[perkam]

We'll have to ask Grayskull to clarify this issue further. I'm sure there are features ATI has that may not necessarily be requirements for SM 3.0 that nvidia doesnt have (like 3dc which only the R420 had) - hopefully there wont be that much of a drop in performance or quality because of it.

Perkam

[Starscream]

i got to say they r right about the: "Shader Model 3.0 Done Right" not being true.
So false advertisement orso.
They do say no current game uses this feature and i doubt many sites will pick up on this but its still bad ATI uses the "Shader Model 3.0 Done Right" slogan if its not true.

im shure Nvidia will pick up on this and wel get a mudd fight. Although i hope not.

edit:
would be cool if there were a way to test this feature and to see how much power this work around drains.

[Vapor]

Could this be what Crytek was talking about when they said SM3.0 on the ATi card need special programming/consideration?

[n00b 0f l337]

Its a let down for you? It says that the feature may not even be used for what its programmed in the current moment. Its no let down. Its like is 7800gtx vs x1800xt, frankly, Its not of the matter to me, as there both too expensive and im an ati fan boy.

[Tim]

Its a let down for you? It says that the feature may not even be used for what its programmed in the current moment. Its no let down. Its like is 7800gtx vs x1800xt, frankly, Its not of the matter to me, as there both too expensive and im an ati fan boy.

Unreal 3 engine is all about Displacement mapping...and that's the first big release next year....

as there both too expensive and im an ati fan boy.

Congrats....

[Cybercat]

I'm sure there are features ATI has that may not necessarily be requirements for SM 3.0 that nvidia doesnt have (like 3dc which only the R420 had)

3Dc is an open standard, and support for it was enabled a long time ago in the Forceware drivers for GeForce 6 and 7 series cards (albiet they do it differently though).

[Vapor]

Yeah, I agree with TIM....ATi's gonna have a problem or two with the Unreal 3 engine.....could Sweeney have known R520 wouldn't have support and have coded around it for X1800 in advance or could this delay the game?

Granted, there is a work around with the Pixel Shaders--but they don't seem to be that good from the preliminary benchmarks we've had.

[J-Mag]

ATi's gonna have a problem or two with the Unreal 3 engine.....could Carmack (being the nVidiot he is) have known R520 wouldn't have support and taken advantage of that?

HUH?

Do you mean Sweeney?

[Vapor]

Uhhhhhh....yeah....sure. I'm bad with names....I'll edit my post now.

[J-Mag]

Uhhhhhh....yeah....sure. I'm bad with names....I'll edit my post now.

heh, i know whatcha mean, but i couldn't tell if you were switching gears in mid thought...

[DilTech]

Tim Sweeney *specifically* told people at E3 that if they're buying a card this generation in anticipation for UT2k7 that they should buy the 7800gtx...

Epic is a 100% NVidia loyal company, the odds of them using this work-around are slim to none... Looks like ATi are in even more trouble.

[Vapor]

Tim Sweeney *specifically* told people at E3 that if they're buying a card this generation in anticipation for UT2k7 that they should buy the 7800gtx...

Epic is a 100% NVidia loyal company, the odds of them using this work-around are slim to none... Looks like ATi are in even more trouble.

I remember that video interview...he said he'd definitely get the G70, but did he know about this (could he have even known about this?) or was it just his bias showing?

[DilTech]

Well, generally, videocard companies tell game developers about features in their upcoming graphics cards in order for the game makers to put them to use in games... I know my boss knew about radiosity in the 7800gtx WAYYYY before the 7800gtx came out, so I'm assuming epic would've had to known... ATi would be RETARDED not to attempt to get a foot in the door on that engine!

Either way, don't expect a work-around for it, epic is one of the more-pro-nvidia companies on the planet!

Finally, I'm assuming he said what he said because radiosity is suppose to be used in the unreal 3 engine, a feature that ATi doesn't even have access to. The guy who wrote that algorithm is also a NVidia fan, as is anyone who ever owned one of the old rage cards. .. This very well could be the reason as well though.

[Turok]

GPU Cloth
This sample demonstrates how to use Shader Model 3.0 to simulate and render cloth on the GPU. The cloth vertex positions are computed through several pixel shaders and saved into a texture. A vertex shader then reads these positions using Vertex Texture Fetch (VTF) to render the cloth.





Simple Vertex Texture
This simple example demonstrates the use of the NV_vertex_program3 extension to perform texture look-ups in a vertex program. It uses this feature to perform simple displacement mapping. The example also shows how to implement bilinear filtering of vertex texture fetches.





Vertex Texture Fetch Water
This sample demonstrates a technique for simulating and rendering water. The water is simulated via Verlet integration of the 2D wave equation using a pixel shader. The simulation result is used by a vertex shader via vertex texture fetch (VTF). The water surface is rendered by combining screen-space refraction and reflection textures.

LINK:
http://download.developer.nvidia.com...les_video.html

Looks like a nice feature
So is this feature hardware dependent?
There is no way to get ATi's drivers, or the game itself patched to support VTF?

[Vapor]

Some good info there Turok...

[DilTech]

LINK:
http://download.developer.nvidia.com...les_video.html

Looks like a nice feature
So is this feature hardware dependent?
There is no way to get ATi's drivers, or the game itself patched to support VTF?

It's a hardware feature, and ATi left it out of their x1800xt, probably by mishap.

Basically, they pulled a NV3x(which left out parts of dx9 IIRC)... it's going to bite them clean in the come UT2k7.

[Tim]

Good info Turok, thanks for sharing

[Ubermann]

Hope someone can test this when they get an XL card.
Is it possible to download this "demos" ?

[perkam]

Basically, they pulled a NV3x

Do you really believe 1500 points more than a gtx in 3dmark05 can constitute "pulling an Nv3x" ??

Yes it lacks one feature, but ATI will tell you they have couple more GTX doesnt have. The GTX doesnt have every graphics feature that has ever been released or ever will be released...it has its drawbacks as well - this is just one for ATI.

I'm hoping ATI clears this up though...there are enough misconceptions about the R5xx line to have ppl claiming its not SM3.0 when they've obviously met at least the minimum requirements for the standard.

Perkam

[RaZz!]

Do you really believe 1500 points more than a gtx in 3dmark05 can constitute "pulling an Nv3x" ??

Yes it lacks one feature, but ATI will tell you they have couple more GTX doesnt have. The GTX doesnt have every graphics feature that has ever been released or ever will be released...it has its drawbacks as well - this is just one for ATI.

I'm hoping ATI clears this up though...there are enough misconceptions about the R5xx line to have ppl claiming its not SM3.0 when they've obviously met at least the minimum requirements for the standard.

Perkam

but this feature is part of an industry programming standard. it's like you're having a (dx9) graphics card which doesn't support antialiasing.
if ati wants to go another way - ok, let them go... reminds me of the time when game developers had to consider different renderers like 3dfx' glide and the powervr-engine...
would be sucky if one day one has to choose a card because of it's different features (and i mean rendering-features...) and not only it's performance.

[Ubermann]

That "one day" came years ago i belive.

[Turok]

All I know is that if this feature eather limits the graphical content or doesnt let me play UT2k7, Gears of War, or any other UE3 games, I may end up buying a nVidia card for the first time.

If the difference is only performance wise, and there is no eye-kandy lost, and it manages to stay above the 7800 GTX in performance, then ATi could still have a chance here.

Im guessing if VTF is really important with the UE3 games, then ATi will probably make a core simmilar to the r520, but just with a minor tweak to add VTF and probably a few more clocks, and then they might release something like the x1_50 line of cards, or they will probably speed up the r580 and add VTF before release.

It would kill ATi if they cant run UE3 games without VTF.
Their cards would be considered obsolete, and nVidia may start making PowerPoint comparisons about it for publicity.
They would crush ATi even more if they decide to release the 7800 Ultra they said was canceled.

[perkam]

Im guessing if VTF is really important with the UE3 games, then ATi will probably make a core simmilar to the r520, but just with a minor tweak to add VTF and probably a few more clocks, and then they might release something like the x1_50 line of cards, or they will probably speed up the r580 and add VTF before release.

I dunno about the VTF in it, but before the R580, a little present called the RV560 will be hitting stores in 2K6

Perkam

[SnipingWaste]

Some points to remember here (and forgive me for the briefness, I'm knocking together a Radeon X1000 technology preview right now which will make mention of this):

- Vertex texturing is not a required feature for Vertex Shader 3.0 support
- Vertex texturing is unusably slow on current NVIDIA hardware
- Vertex texturing requires a large amount of transistors and silicon to be implemented properly without the move to a unified shader architecture
- ATI are offering 'Render to Vertex Buffer' as an alternative to vertex texturing which can be utilised using a FOURCC identifier. This will offer the same functionality as vertex texturing, but with better performance

Heres more info