How easy is it for ATI to beat Geforce FX?

[StealthHawk]

where was that quote from? i'm not caught up in all the latest NV30 news, about 1 day to 1/2 day behind.

[Bigus Dickus]

It was taken from a discussion about displacement mapping in the NV30 on Beyond3D's forums. A member, tb I believe, asked Doug the question (presumably through email, though I'm not certain of that), and that was Doug's response.

It could be completely fabricated, or a simple misunderstanding.

[thcdru2k]

whats displacement mapping :D

any games planning to use it?

[Nv40]

yeah! While ATi says on paper that its support DIsplacement
mapping it is not even clear if ATi support "True" displacement
mapping like Matrox . Beyond3d there is a very interesting topic there ;)


off topic..

for Antialiasing lovers , they will be happy to know that
Nv30 antialiasing improved techniques looks finally
outstanding as good if not better than any others gaming cards in the market :)


http://www.nvidia.com/docs/IO/3666/SUPP/large01.jpg

http://www.nvidia.com/docs/IO/3666/SUPP/large02.jpg

http://www.nvidia.com/docs/IO/3666/SUPP/large03.jpg

http://www.nvidia.com/docs/IO/3666/SUPP/large04.jpg

Wow! The first time i saw an screenshot of that game
in Nvnews front page ,i thought it was a Photo of a real
car competition .

and even True -128bits colors precision as a Bonus!!
at playable framerates :)

http://www.nvidia.com/docs/IO/3652/SUPP/large01.jpg

http://www.nvidia.com/docs/IO/3652/SUPP/large03.jpg

http://www.nvidia.com/docs/IO/3666/SUPP/large04.jpg

eye candy to the Max! :)

for more screenshots here ..

http://www.nvidia.com/view.asp?PAGE=geforcefx_games

[Bigus Dickus]

That AA didn't blow me out of my chair by any means, but it would probably be hard for a .jpg with no reference to do so anyway. :rolleyes:

And I seriously doubt the game screenshots at the bottom used 128 bit floating point color.

[Fotis]

Originally posted by thcdru2k
whats displacement mapping :D

any games planning to use it?

As you can see I currently own a gf2 and it does not support EMBM because nvidia disided not to support it a couple of years ago but I think thats not the case this time because their official site mentions displacement mapping.

[tamattack]

Originally posted by Fotis
I think thats not the case this time because their official site mentions displacement mapping.

That may be true, but we have seen NV post 'features' to their website, only to remove them later.

Example: NV website used to claim HOS (RT splines?) support for GF3/4. Then one day they removed support from the drivers and removed the 'feature' from their website.

This is not to say that other IHV's have not done so in the past as well. Just that it's a little premature to point to this as 'proof' when no one has been allowed to independently put the claims to the test.

[tamattack]

Originally posted by Joe Cool
Waiting is always better if you can afford to, since something better will always eventually come along, in the computer business.

IMO waiting is never better. As you say, something better will always be on the horizon. Following your logic, you would never actually upgrade, since you would always be waiting for the next big thing.

Originally posted by Joe Cool
Is it [GFFX] enough better to be worth waiting for?

That is THE question!

Originally posted by Joe Cool
But waiting for the R350 when we have so little info on it? Doesn't make any sense to me....

No different that waiting for the NV30 for the past 4 months (prior to Comdex) when there was so little info about it.

So the question I have for you is: 'Why did you wait for NV30?'

[tamattack]

Originally posted by thcdru2k
the doom 3 benchmarks could be hashed, the graphs on that site are all homemade. however, the nvidia benches, that consisted of doom 3, ut2k3, quake 3, nature test, etc where all real as democoder of b3d forums had said they had the unreal benchies running at comdex for over two hours. yes, the r350 will be faster. the race is so tight ati would never release anything slower. its all about when, and how nvidia responds, and how ati responds, etc.

I could be wrong, but I don't think Democoder ever said that he witnessed the benchmarking. IIRC he said he witnessed a number of demos/games running in real time on actual hardware. There's quite a distinction (again, IIRC).

[tamattack]

Originally posted by gstanford
ATI would have to implement 128 bit floating point color right through the entire pipeline - that means a serious redesign of R300 which only uses 128 bit in certain places and only calculates to 96 bit in the shaders.

Why would you say that? Up to this point, no one has been able to show that 128 bit FP is noticeably superior to 96 bit FP. Although I don't doubt that it is incrementally better, the true revolution is in the move from integer to FP.

In fact, IIRC, Carmack's wishlist was for 64 bit FP, and both ATI and NV have surpassed that.

[tamattack]

Originally posted by gstanford
Who cares about DirectX9.0? It's DirectX9.1 that will matter

ATI supporters said the same thing back in the DX8 vs. DX8.1 debate. That has not turned out to be the case. Why should things be any different now?

And what side of the fence were you on during that debate?

[tamattack]

Originally posted by Nv40
yeah,, and it is funny too , that Ati a "direcx9 card"
was finished many months before DirectX9 beta features
were Final !!!!

supporting much less Nv30 Directx9 Vs/ps features and still claim "full support" for directX 9. ;)

R300 is a full DX9 part. It was MS' target/reference platform for DX9 dev.

NV30 has been positioned as DX9+ (but not full DX9.1).

[tamattack]

Originally posted by Uttar
LOL, yeah, sounds like that was a joke :)
However, the document I linked to isn't outdated. It was released by nVidia the two days ago.
Let me link it again in case some people didn't see it:
http://www.nvidia.com/docs/lo/2413/...ders_110402.pdf

It clearly proofs the GeForce FX supports Displacement Mapping.

There's no proof until hardware is available for people to put it to the test.

Remember RT spline-based HOS? It too was on the NV website until one day NV disabled it in the drivers and removed it from the website.

[Fotis]

Additional info on r350:
http://digitimes.com/NewsShow/Article2.asp?datePublish=2002/11/20&pages=08&seq=42

It looks like ATI is waiting to have samples of its new 0.13 card back from the fab by years end.It looks interesting to say the least!!

[Dazz]

Originally posted by Dr_Colossus
I also have doubts about how well ATI can ramp up there architecture in the case of the R350, I wouldn't expect much considering the GFFX only uses 17 million more transistors and delivers incredible numbers in terms of features. I expect the R350 to use more transistors to accomplish the same tasks as the GFFX. You forget the Radeon 9700 can hit 400MHz on 0.15m and with simple quiet cooling.

[thcdru2k]

i have one thing to say: nvidia is gonna lose a lot of ground on ati if they don't get that nv35 out right after the r350.

[ChrisW]

You guys do realise what you are calling 128bit color is not something that is going to be displayed, right? Currently, your display uses 24 bit color (16.7 million colors) of which the human eye can only see a little over 7 million different colors (not to mention your monitor can only display 24 bit color). 128bit color would also require four times the amount of video memory to store (which would slow your computer to a crawl unless you had an incredible amount of video memory). What is being refered to as 128 bit color is nothing more than floating point registers. It is used to do logical calculations on a color bit. When all the calculations are completed, it is reconverted back to 24 bit color and then displayed. The only difference in the final image will be that some pixels were rounded up or down to the nearest color during the calculation without using the higher precision registers. Yes, 128bit precision is better (more accurate) but you have to realise it is an astronomical number (just like a 96 bit number). I can guarantee if you were shown three images, one using 128bit precision, one using 96bit, and the third using 64 bit, you would be hard pressed to tell the difference.

[nutball]

Originally posted by ChrisW
You guys do realise what you are calling 128bit color is not something that is going to be displayed, right?


Yes.


Currently, your display uses 24 bit color (16.7 million colors) of which the human eye can only see a little over 7 million different colors (not to mention your monitor can only display 24 bit color).


Analogue monitors are capable of displaying more than 16.7 million colours. As for the human eye bit, it's not necessarily the total number of colours which is the issue. The human eye can perceive more than 256 shades of one colour. Look at the graduated blue-ish screen you get when installing software. The steps in the colour are clear. It's an effect called Mach banding.

128bit color would also require four times the amount of video memory to store (which would slow your computer to a crawl unless you had an incredible amount of video memory).

These cards will have an incredible amount of memory (128-256MB), and the bandwidth to go with it.


What is being refered to as 128 bit color is nothing more than floating point registers. It is used to do logical calculations on a color bit. When all the calculations are completed, it is reconverted back to 24 bit color and then displayed. The only difference in the final image will be that some pixels were rounded up or down to the nearest color during the calculation without using the higher precision registers. Yes, 128bit precision is better (more accurate) but you have to realise it is an astronomical number (just like a 96 bit number). I can guarantee if you were shown three images, one using 128bit precision, one using 96bit, and the third using 64 bit, you would be hard pressed to tell the difference.

What you say is not entirely true.

Firstly it's true that the 128-bit-per-pixel modes aren't available for the frame-buffer. They are available for off-screen buffers. You can render into them, and the pixel component values are stored as 32-bit floating-point numbers, they are not converted to 8-bit integers.

Secondly, well. Blending, blending, blending. The 128-bit modes offer two things: more precision and higher dynamic range. This is the key. It really depends on what you're doing, but I can guarantee that there are applications out there which will benefit from floating-point pixel formats. I know this, because I'm writing one!

[ChrisW]

I agree, applications will definately benefit from floating point color precision. I just think the difference between 64, 96, and 128 bit precision will be small.

Analog monitors may be capable of displaying more colors, but not digital monitors.

As far as having plenty of video memory, if games are loading 180MB of textures, you don't have plenty of memory.

[pastor]

the goal of 128 bit precision (or whatever 96) has never been targeted to be displayed on screen

it's a matter of internal precision rendering

For instance, one benefit will be to use HRDI (high dynamic range image) for textures wich has the advantage to give to 3d scenes a more realistic appearance thx to the wider range of colors values

this exist already for traditional pre rendered 3D:


Rendering without HRDI environnement map

http://www.debevec.org/HDRShop/main-pages/images/memorial2.jpg

Rendering with HRDI environnement map

http://www.debevec.org/HDRShop/main-pages/images/memorial1.jpg

(the motion blue effect has no direct relation with HRDI rendering in this scene, however HRDI texture improve a lot the visual quality of motion blured renderings )