Anyone else wondering if 6800 Ultra is really the 6800?

[Joe DeFuria]

By ATI, or the developer? And isn't that particular feature now being touted by nVidia as well?

It appears that this more a vindication of Valve's stance on MRT, than it is ATI bashing nVidia for a feature ATI "claims" is significant.

That might be correct. But Valve and ATI has pretty strong bonds though.

Correct...but if we're to believe Valve, the bonds are strong because ATI had the better hardware implementation. So we're back to a software dev evangelizing a feature from an IHV that they feel is important...that in the next gen the competing IHV implements and specifically touts as useful....

 

[Joe DeFuria]

The real question is whether or not PS 2.0 will become a deflection point with game developers. I think nVidia certainly wants it to be, what with their NV3x's poor performance with 2.0 shaders and their support of the 3.0 model with their NV4x parts. But I think Intel's new integrated graphics and the market share of ATI's R3xx chips won't allow it. Regardless, it's a shame ATI doesn't support 3.0, just as it was a shame nVidia's FP performance has been so poor, pre-NV4x.

Agreed on just about everything. We'd all prefer if ATI did support PS 3.0, but that doesn't change the fact that PS 2.0 is emerging / has emerged as the baseline standard.

 

[Stryyder]

By ATI, or the developer? And isn't that particular feature now being touted by nVidia as well?

It appears that this more a vindication of Valve's stance on MRT, than it is ATI bashing nVidia for a feature ATI "claims" is significant.

That might be correct. But Valve and ATI has pretty strong bonds though.

Correct...but if we're to believe Valve, the bonds are strong because ATI had the better hardware implementation. So we're back to a software dev evangelizing a feature from an IHV that they feel is important...that in the next gen the competing IHV implements and specifically touts as useful....

Which I am sure John Carmack will be doing about NVIDIA and NV4X Ultra Shadow II and PS 3.0 support unfortunately

 

[Bjorn]

The real question is whether or not PS 2.0 will become a deflection point with game developers. I think nVidia certainly wants it to be, what with their NV3x's poor performance with 2.0 shaders and their support of the 3.0 model with their NV4x parts. But I think Intel's new integrated graphics and the market share of ATI's R3xx chips won't allow it. Regardless, it's a shame ATI doesn't support 3.0, just as it was a shame nVidia's FP performance has been so poor, pre-NV4x.

I don't think that it's going to be a deflection point either. Although Carmack is going to use the NV40 when developing his next engine so who knows (doesn't mean that it'll be the minimum required though, the NV40 might be very old news when that engine is released)

And afa Intel's integrated graphics goes, well, i'm guessing on DX9 performance that's so low that it'll end up running low end DX8 paths but we'll see about that of course.

 

[Stryyder]

The real question is whether or not PS 2.0 will become a deflection point with game developers. I think nVidia certainly wants it to be, what with their NV3x's poor performance with 2.0 shaders and their support of the 3.0 model with their NV4x parts. But I think Intel's new integrated graphics and the market share of ATI's R3xx chips won't allow it. Regardless, it's a shame ATI doesn't support 3.0, just as it was a shame nVidia's FP performance has been so poor, pre-NV4x.

Agreed on just about everything. We'd all prefer if ATI did support PS 3.0, but that doesn't change the fact that PS 2.0 is emerging / has emerged as the baseline standard.

Even if it is the 'baseline standard' NVIDIA will actively pursuing the developers to 1)Add it to games already offer 2.0 support and 2) make sure all new games feature it.

If it is allegedly so much easier to implement than 2.0 for the developer as all these tech articles are saying, what is to prevent them from being successful and therefor trumping ATI and their lack of PS 3.0 support?

 

[Joe DeFuria]

If it is allegedly so much easier to implement than 2.0 for the developer as all these tech articles are saying, what is to prevent them from being successful and therefor trumping ATI and their lack of PS 3.0 support?

Who said it was easier to code for while at the same time maintaining performance?

 

[Bjorn]

If it is allegedly so much easier to implement than 2.0 for the developer as all these tech articles are saying, what is to prevent them from being successful and therefor trumping ATI and their lack of PS 3.0 support?

That of course assumes that there is a noticeable difference when going to SM 3.0 (speed, IQ or cpu usage)

 

[Stryyder]

If it is allegedly so much easier to implement than 2.0 for the developer as all these tech articles are saying, what is to prevent them from being successful and therefor trumping ATI and their lack of PS 3.0 support?

Who said it was easier to code for while at the same time maintaining performance?

Every 6800 Ultra preview I have read so far today

from Hard OCP

Advancements in CineFX’s technology have raised the capabilities of pixel shaders so they’re on par with vertex shader capabilities. With increased control at the pixel level, programmers can dramatically enrich their games, bringing more lifelike qualities to every character, object, and scene. The native 32-bit processing abilities of the GeForce 6 Series of GPUs also increase overall pixel shading precision, taking pixel shaders to new levels of unsurpassed image quality."

Just as there are new features with Vertex Shader 3.0, there are new features with Pixel Shader 3.0.

"No longer forced to restrict each pixel shader program to 96 instructions, programmers are now free from hardware2 limitations, and can implement more complex effects at the pixel level."

Just like the Vertex Shader, the Pixel Shader’s length is infinite. The shader program limit for PS 2.0 was 96 instructions; with PS 3.0 on the NV40 it is infinite.

"Full support for subroutines, loops, and branches—including loop counter registers and condition codes—and a new back/face register gives complete control to the programmer."

The same flow control capabilities of loops, branches, and subroutines are also supported in the pixel shader.

"The GeForce architecture has always enabled developers to choose the necessary level of precision for each image or scene. Now the choice is simpler, because the performance degradations associated with full 32-bit floating-point precision have been eliminated."

"Floating point operands can be treated in native 32-bit or optional 16-bit format, which are the standard formats in the film industry. Although both modes deliver equivalent performance, the 32-bit floating point mode uses twice as much memory to store the operands. Programmers can choose between native 32-bit mode and the optional 16-bit mode to achieve the required level of precision in each case. Plus, they can efficiently manage memory usage in situations where space is a consideration. Other data formats are also supported."

In DirectX9.0, Floating Point 24 was required as the minimum value for PS/VS 2.0. For PS/VS 3.0 the minimum requirement is FP32.

Floating Point 32 (FP32) is supported across the entire pipeline in NV40. The NV30 series also supported FP32. However, performance of FP32 in the NV30 series was very slow and often precision was lowered to gain performance in games. With NV40, NVIDIA claims that performance has been upgraded and is now very fast. In fact, above it states FP32 performance and FP16 performance are “equivalentâ€￾, though keep in mind FP32 uses twice as much memory to store operands.

 

[Joe DeFuria]

Every 6800 Ultra preview I have read so far today

Um, Again, where does anyone say or imply that it's easier to code for PS 3.0, and that PS 3.0 means better performance? Certainly nowhere in that blurb you postsed.

All I see a lot of architectural improvements in NV40 (that apply to both PS 2.0 PS 3.0 performance) compared to NV35.

 

[Stryyder]

Every 6800 Ultra preview I have read so far today

Um, Again, where does anyone say or imply that it's easier to code for PS 3.0, and that PS 3.0 means better performance? Certainly nowhere in that blurb you postsed.

All I see a lot of architectural improvements in NV40 (that apply to both PS 2.0 PS 3.0 performance) compared to NV35.

No longer forced to restrict each pixel shader program to 96 instructions, programmers are now free from hardware2 limitations, and can implement more complex effects at the pixel level."

ust like the Vertex Shader, the Pixel Shader’s length is infinite. The shader program limit for PS 2.0 was 96 instructions; with PS 3.0 on the NV40 it is infinite.

"Full support for subroutines, loops, and branches—including loop counter registers and condition codes—and a new back/face register gives complete control to the programmer."

The same flow control capabilities of loops, branches, and subroutines are also supported in the pixel shader.

Unlimited number of instructions, infinite length and full support for subroutines, loobs, and branches all seem to imply improvements over 2.0 but again I am not a coder, I guess a developer would have to give their two cents.

 

[John Reynolds]

You have to understand that most of these previews are simply regurgitating the marketing material that was handed out along with the boards. Which means they contain a combination of facts and some manipulating of the facts to put your new product in a better light.

 

[Stryyder]

You have to understand that most of these previews are simply regurgitating the marketing material that was handed out along with the boards. Which means they contain a combination of facts and some manipulating of the facts to put your new product in a better light.

Is that a nice way of saying that the PS 3.0 could possibly be no better than ps 2.0 and the only 'advantage' is that is is FP 32 which plays up NVIDIA architecture??

 

[Joe DeFuria]

No longer forced to restrict each pixel shader program to 96 instructions, programmers are now free from hardware2 limitations, and can implement more complex effects at the pixel level."

So longer shaders are now faster than shorter ones?

Just like the Vertex Shader, the Pixel Shader’s length is infinite. The shader program limit for PS 2.0 was 96 instructions; with PS 3.0 on the NV40 it is infinite.

How's the performance looking on those infinitely long shaders?

"Full support for subroutines, loops, and branches—including loop counter registers and condition codes—and a new back/face register gives complete control to the programmer."

The same flow control capabilities of loops, branches, and subroutines are also supported in the pixel shader.

And how fast is the hardware at performing these operations, branching, etc,

(etc., etc.)

Do you get the idea? You can repeat yourself and bold all you want...it doesn't change the fact that this does not address the question you claim has been answered.

 

[Joe DeFuria]

You have to understand that most of these previews are simply regurgitating the marketing material that was handed out along with the boards. Which means they contain a combination of facts and some manipulating of the facts to put your new product in a better light.

Is that a nice way of saying that the PS 3.0 could possibly be no better than ps 2.0 and the only 'advantage' is that is is FP 32 which plays up NVIDIA architecture??

First of all, no one is denying that PS 3.0 isn't "better" than 2.0. It is.

The question is to what meaningful extent. For what practical extent is coding for PS 3.0 going to give over a PS 2.0 shader for this generation of hardware?

FP 32 is NOT AND ADVANTAGE of PS 3.0 per se. nVidia supports FP 32 in PS 2.0.

The "big deal" about nvidia's shaders in the NV4X is the big improvement in performance and apparent robustness of 2.0 at "full precision." PS 3.0 is icing on the cake, but is not why I appreciate the NV4x architecture.

 

[Stryyder]

No longer forced to restrict each pixel shader program to 96 instructions, programmers are now free from hardware2 limitations, and can implement more complex effects at the pixel level."

So longer shaders are now faster than shorter ones?

Just like the Vertex Shader, the Pixel Shader’s length is infinite. The shader program limit for PS 2.0 was 96 instructions; with PS 3.0 on the NV40 it is infinite.

How's the performance looking on those infinitely long shaders?

"Full support for subroutines, loops, and branches—including loop counter registers and condition codes—and a new back/face register gives complete control to the programmer."

The same flow control capabilities of loops, branches, and subroutines are also supported in the pixel shader.

And how fast is the hardware at performing these operations, branching, etc,

(etc., etc.)

Do you get the idea? You can repeat yourself and bold all you want...it doesn't change the fact that this does not address the question you claim has been answered.

The bold was used to seperate your reply which was quoted and the specific information I was extracting from previous post. Why you are turning this into a confrontational situation when all I am doing is asking questions based on the impressions I am getting as a layman from these posts and articles and history. I specificaly said in my post that what was stated IMPLIES that PS3.0 is easier to code than 2.0 and that I don't understand it completely. Subsequently someone else pointed out that this was probably just restating NVIDIA PR Babble and that it should be taken with a grain of salt.

You sir seem to be defensive for no descirnable reason and bordering on being offensive for no reason. People come to these forums because they have questions, which is all I am doing, if you don't like my questions, or the reasons I am asking them or how I am interperting the data you can attempt to educate me with the facts that you have rather than question my questions and be offensive.

I am an ATI fan and have only owned ATI cards since the Rage Fury Pro I hade a few years ago to the 9600xt I own today, you sir seem to be a FanATIc and I am just now understanding what that means.

 

[KimB]

First of all, no one is denying that PS 3.0 isn't "better" than 2.0. It is.

The question is to what meaningful extent. For what practical extent is coding for PS 3.0 going to give over a PS 2.0 shader for this generation of hardware?

I still say the biggest difference is not in the shaders themselves, but in the FP blending/filtering.

As a side note, I'm really, really excited that the hardware has unlimited length in the shader programs. I'm hoping that this will be exposed in GLSL on launch.

 

[Bouncing Zabaglione Bros.]

Every 6800 Ultra preview I have read so far today

from Hard OCP

You realise this is all the marketing blurb put out by Nvidia for these preview events and copied verbatim by the websites?

 

[Joe DeFuria]

You sir seem to be defensive for no descirnable reason and bordering on being offensive for no reason.

No, I get that way when I have to repeat myself three times, and the response I get from you was no different than the first two times I told you that what you said doesn't address your point.

 

[DemoCoder]

I think people are focusing too much on the extra instructions in PS3.0. I think the big news for SM3.0 in the NV40 is

FP32 seems to run adequately
FP filtering and blending
Vertex texturing
Geometry instancing (may be possible on some VS2.0 cards, unknown yet)


The untold story of the NV40 is the video processor. If you look at the diagrams, this thing is effectively a Scalar+SIMD semi-general purpose CPU/DSP on-core running at a different clock rate, with it's own memory controller and read/write memory heap. I think this thing can do alot more than video encoding/decoding and is an untapped potential. Think "video processor -> render to vertex buffer" and physics simulation, or audio simulation.

 

[Johnny Rotten]

Whether the absence of SM3.0 in ATI's next gen hurts them will, as always, be left to the developers. Twice ATI has had superior shading capabilities vs nvidias competing line: R200 vs nv2x and r300/r350 vs nv3x, only once has that advantage 'hurt' nvidia. If applications (games) can provide a demonstrable advantage under the SM3.0 model (either from an IQ or performance standpoint) then obviously ATI is going to be competing from a position of weakness. There are some fairly high profile titles though having or promising SM3.0 support though (Far Cry, STALKER, HL2) and apparently the difference in Far Cry (based on a 3rd person impression mind you) at least is significant.