NV30 Rumors, a big bad boy, Matrox is in trouble (maybe :) )

[CaineTanathos]

Some urls:

http://www.sharkyextreme.com/hardware/videocards/article.php/1434621

http://www.nvmax.com/Articles/Previews/NV30_SNEAK_PREVIEW/

Rumored specs (what I can make out of it) :

.13 micron
agp 8x
GPU will run a at a speed of 400 mhz
256 bit memory bus
8 Pixel units
each pixel unit 2 texture units
16 textures per pixel in one pass


internal use of use of 64 and 128 Bit Color rendering
supports full 16- and 32-bit floating point formats (for colors, its proberly the same as that 64 and 128 bit color )

The nv30 will support 12 bit fixed point, 16 bit float, and 32 bit float
-- So 48bit color support (they proberly stole it from 3dfx napalm )


NV30 is different from its predecessors because it focuses more on the GPU rather than the memory. The shift means that the performance bottleneck shifts from memory efficiency to computing efficiency. In addition to this the NV30 includes better algorithms for detecting when hidden pixels that applications are not rendered through the system. This is known as an extension to Hidden Surface Removal and will be a part of a new system NVIDIA said we couldn't talk about, at this time.

-- Does Matrox use algorithmes ??

-- I am still wondering why matrox didn't implement PS2.0 and support for floating point formats for the colors (but thats a part of PS2.0)



Carmack quote:
Nvidia is the first of the consumer graphics companies to firmly understand what is going to be happening with the convergence of consumer realtime and professional offline rendering. The architectural decision in the NV30 to allow full floating point precision all the way to the framebuffer and texture fetch, instead of just in internal paths, is a good example of far sighted planning. It has been obvious to me for some time how things are going to come together, but Nvidia has made moves on both the technical and company strategic fronts that are going to accelerate my timetable over my original estimations


oh yeah :
Annoucement date august/november
and you proberly can buy one around christmas - february
well I am guessing am I

[TdB]

I don´t think matrox will get in trouble because of this, its a different market, nvidia doesn´t focus on the same things as matrox.

[CaineTanathos]

proberly,

I still am gonna buy a parhelia though

[Drizzt]

Ok, if this is true then nVidia's engineers are the best engineers in the world.

A chip supporting all the above fp maths and the normal features of directx-opengl supported by P and R300 must have, at a glance, almost 200 mln of transistors.
Those 200mln run at 400mhz, so Nvidia must have discovered something really interesting to keep low the heat.

So, if this is a consumer card, Nvidia must have discovered the way to manufacture such a product at really low price. Or, the card will cost over 1000$...


Bah...

[CaineTanathos]

its made in 0.13 micron so thats cheaper too
I didn't find anything about the actual number of transistors
But I think it will be more around 150 million

[Electric Amish]

its made in 0.13 micron so thats cheaper too

How's that??

That also depends if they ever get it out.

Rumours have it that they're having .13u problems and it won't be out until next year.

amish

[Greebe]

Whenever anything is first brought to market, especially something special like a fab that will have 130nm process they charge the most for it.

TSMC (nV's current fab) is having problems refining their process atm. I bet nV will not have this in a sellable product until at least Q1 '03.

[CaineTanathos]

Can anyone tell me if 2x , 4x , 8x anisotropic filtering is the same as 16-tap , 32-tap, 64-tap anistropic level?

yeah, yur proberly right, but maybe they are using a secret fab
I am also wondering if they get 12 bit color ramdacs

I am still getting a parhelia, but I am gonna wait till the opengl speed is fixed, because at the moment my GF3ti 200 is faster even at normal speed.

But I am sure it will be.
fixing the 2x tap filtering and adding 4x and 8x would be nice.
Matrox also claimed that with current games like Counter-strike and max payne , quake3 , etc without permance loss.

But again I bet they will fix it in future driver updates (I hope, no I beg, PLEAAAASE )

[Drizzt]

Caine etc., to handle 128 bit internal color rendering you do need 4 times the transistor needed for 32 bit rendering.
And apart from the Vertex Engine, everything else on the GPU does color computation...

[CaineTanathos]

but its only for the textures , so not the whole line
its 48 bit for the whole chip (so 12 bit per color, also alpha) not sure though, but wouldn't 3dfx had that too if they had released their napalm card.
so 48 bits is a 50 % increase so 150 - 200 m transistors is
proberly right
Pixelshader 2.0 has a seperate color processing unit , this is the place where it uses the 64bit/128 bit floating/integer colors

Its weird that videocard makers never used floating point for colors, because its much more accurate than integer, but its proberly a new step in the graphics evolution.




But its also normal that the parhelia doesn't support that many bits per color because its pixelshader is only version 1.3 so it doesn't support more than 4 textures per pass (thats version 1.1, but I don't know if something changes with version 1.3), so the texture corruption is much less.

And you proberly need software support for enabling 64/128 bit colors internal (textures , since pixelshader 2.0 can handle 16 textures per pass).

still stupid that matrox doesn't have 10 bits alpha channel support or 8 bits when using 10bits per color. Maybe it does, because Matrox said they support the 10:10:10:2 format, so maybe if there will be a 10:10:10:10 format they can support it too.
But if they don't ,well I think they did a good research before making this chip.

can anyone answer my anisotropic question

[Novdid]

So it seems the only way nv could counter ATI was to release some techsheets, that creates rumours.

Example: If ATI can get 300MHz with .15 then nv can get...hmm...say 400MHz with .13.

[Ssin]

Hey maybe they found a way to put micro fans into the chip... damn I want that in my cpu.

[Rakido]

You mean that?

"Fans the Size of a Grain of Sand May One Day Cool Computers"
New York Times (02/15/01) P. E10; Eisenberg, Anne
Researchers at the University of Colorado at Boulder used solder to create a microfan that hoists fan blades up from the chip, compared to the current acid bath technique of burning objects out of chips like sculpture from stone slabs. More than 3,000 microfans can cover 1 cubic inch, and researchers are now eying them as chip-level cooling tools for personal computers. The first blade was constructed by mechanical engineering graduate student Paul E. Kladitis, who was able, through this technique, to create fan blades at an angle: like upside-L shapes with the small foot at any desired angle. The old acidic method could only create objects within the chip's plane, while the new solder method can pull blades up at an angle to overhang the chip's surface. Dr. Richard R. A. Syms of Imperial College, London, has been continuing work on this promising method. "We have demonstrated it with meltable pads of solder, glass or polymer, rotating parts made of silicon, copper or nickel," Syms says. However, more engineering must be done to see if microfans can run at necessary speeds and whether they possess necessary strength, among other questions.

Rakido

[borat]

yes and also matrox will be rleasing on .09 micron not long after while ati move straight to .07 micron and the race is on to put 1000 million transistors on a gpu and bitboyz have uncomfirmed runours that they have a 4ghz gpu ready for launch a few days after ati. we have heard all this hype before, yeah nv30 will undoubtedly be fast but its 6 months away minimum and will cost a fortune and will be eclipsed 6 months after by some other company whoever they may be and the official spec will only be 75% of the rumoured spec and the darn thing will still be blurry. i am no matrox fanboy and probably wont even get parhelia due to cost/performance but please lets just treat these rumoured specs with a pinch of salt and look at what we have now ie parhelia radeon 9700 and gf4ti if you are obsesed with future cards you will never by any because there is always another round the corner. think back to this time last year we had gf3 then radeon 8500 then gf4 then we waited for parhelia then radeon 9700 came along each time one was released another was around the corner.

[TdB]

some of us, actually like these speculation-threads.
I like to speculate about future cards, but that doesn´t mean I will buy them all.