[muted]

http://forums.matroxusers.com/showth...threadid=32858

well .. speaks for itself..

looks nice ..

[nAo]

Umh..Triple head, 80Mtransistors, 256 bits wide DDR bus...shaping nice

[Humus]

Tuesday is certainly going to be a day to remember.

[Tatchan]

How brave they are! Just wonder if Matrox will do anything on this...

[nggalai]

Yes, so much for NDAs . . . I mean, really.

...

...

Not as if I was complaining. Triple-screen FS looks nice, and if they show it off, it means that the framerate is more than ok, too. But, eh. Speculations, we'll see on Tuesday. But the comments of the two mods regarding FSAA, or rather the lack of comments and amount of smileys, makes me fear for my wallet . . .

ta,
.rb

[maskrider]

This year is really very interesting, and will definitely hurt my wallet more than ever.

http://hard.zol.com.cn/2002/0511/40598.shtml

"Parhelia是世界上第一款512位的GPU，采用0.15微米生产工艺，集成了8000万颗晶体管 ，具有256位的显存，显存带宽超过20GB/s"

Parhelia is the world's first 512bit GPU, which is fab on the 0.15 process, and contains 80M transistors, the 256bit memory bus provides more than 20GB/s bandwidth.

[nAo]

Originally Posted by nggalai

But the comments of the two mods regarding FSAA, or rather the lack of comments and amount of smileys, makes me fear for my wallet . . .

8x..16x..usable..coughh....fsaa...

[Fuz]

I am glad this tuesday won't be just a 3d technology preview like 3dfx had with voodoo4/5. Its good to see that they actually have working hardware, even if it many not be ready for prime time (production). Acutally am a bit surprised.

Really looking forward to it, its good see competition back in the industry...

I might have to buy one, tripple head looks awesome! Forget games.... porn on 3 monitors at once!!!

[pascal]

The second chip with 256bits DDR memory bus.

Imagine very large outdoors all DOT-3, great lighting, lots of polygons, nice filtering and high fidelity 3D.
What will they do with .13 micron? Maybe the same Parhelia at lower price with different bandwiths for different markets.

I hope Matrox will forgive them because it is really exciting.

[Grall]

Does anyone have the thing cached? It seems to be gone now...

*G*

[Onslaught]

I just can't stop staring at those triplehead pics. That's awesome! I knew they could do it. I'm glad I have the money ready for the high-end gamers version from day one.

[nggalai]

Hi there,

Imagine very large outdoors all DOT-3, great lighting, lots of polygons, nice filtering and high fidelity 3D.

Yes. In something like 2-3 years perhaps . . . as long as the common denominator can't cope with that . . . but we've had this discussion before, so I'll stop.

Not to make even more room for speculation here, but regarding the actual shipping of the hardware: I've "heard" from three different, unrelated sources that we're not talking about an autumn date, but more something on the lines of the next couple of weeks / few months . . . let's hope for the best. (And don't forget that's all speculation until Tuesday.)

@nAo

8x..16x..usable..coughh....fsaa...

Argh, stopitstopitstopit. OK, there goes my holidays on the Bahamas for this year . . .

ta,
.rb

P.S. I apologise if I come over too much like a fan fag. But the idea of a Matrox board that's more than decent for actual gaming is making me quite . . . happy. Yes, happy. Er.rb

[pascal]

Yes. In something like 2-3 years perhaps . . . as long as the common denominator can't cope with that . . . but we've had this discussion before, so I'll stop.

Ok, I will be happy to have a 1024x768x40 highly filtered stochastic FSAA at 60fps with Doom3 this year.

[Tagrineth]

Jeeze, talk about exponential performance increases lately.

Way back when, it was a question of 30-50 FPS between generations (for the most part)... now we're talking 100-130 more FPS...

You know, I just realised something:

This jump may actually beat the legendary Voodoo Graphcs -> Voodoo2 Triple-performance gap.

[Kristof]

fab on the 0.15 process, 80M transistors, 256bit memory bus

This screams hot and expensive. 80M transistors in .15 is going to be quite big, and silicon area = costs, not to mention "hot" when being run at reasonible clockspeed... but impressive.

Tripple Head... kewl on paper but even less useful and realistic than Dual Head. Having 2 display devices on your desk : maybe, but 3 ?!

[nggalai]

Hi Kristof,

my thoughts exactly. .15um and a gate count of 80Mio . . . eeek. This thing will be VERY pricy.

Regarding your argument againts triple-head: I see it like pascal above--if the thing has enough power to drive three screens, I'll be happy using that juice for some nice texture filtering and FSAA . . .

ta,
.rb

[pascal]

About price, TSMC .15 micron is old and some 53m transistors .15 micron GF3Ti200 today cost $100.

I will not be suprised if they sell it for $400.

[pascal]

OH my god http://ixbt.com/video2/matrox-parhelia512.shtml
Got the link from MURC.

[CoolAsAMoose]

My Russian is not that good, but a few interesting details can be found:

80 million transistors on 0.15 um process
350 MHz (?)
256 bit memory IF
4 vertex shaders (2.0)
4 pixel pipelines with 4 texturing units each, pixel shader v1.3 (although they are talking about 2.0 as well)
16x AA
2 integrated RAMDACS, digital out can connect to 3:rd RAMDAC

And, of course, displacement mapping. That Grand Canyon looks cool!

[Humus]

Now this chip really seams very cool

[MfA]

WOW Pascal, just gonna think out loud a bit ...

Assuming those 128 bit vertex shader engines do 1 MAC per cycle thats 350 MP/s.

I see the term anisotropic filtering wasnt fancy enough for Matrox :)

The anti-aliasing looks very cool, quality wise its exactly (!) the same as 16x multisampling as long as the fragment buffer doesnt overflow AFAICS. Although it wont AA intersections, so its more like a modified A-buffer approach.

[L233]

Let's see what the Babelfish says...

- 80 million transistors
- 0.15 process
- chip and memory clock up to 350 MHZ
- 256 bit BUS
- 20 GB/s bandwidth
- 8x AGP with SBA and Fast Writes
- 4 pipelines with 4 texture units each
- Vertex Shader 2.0, four parallel shader units, some DX9 stuff supported
- PixelShader 1.3... 4 texture + 5 combination stages on each pipeline (?)
- possibility to combine the pipelines in pairs for 10 combination stages (?)
- EMBM and DOT3
- some sort of fixed T&L support with extended matrix blending and skinning or something
- 400 MHZ, 10bit RAMDAC and something about 10bit TV-Out thingy
- triple head (kinky!) support for surround gaming
- 16x "Fragment AA" only applied to boundry points of polygons or whatever (?)
- No clue what this means but there's an exclamation mark so I guess it's important: "To glypyu Of antialiasing.the technology of apparatus boundary smoothing and gamma- correction of types (!)"
- N-Patches with adaptive tesselation (?)
- displacement mapping
- DX8/OpenGL 1.3 and some DX9 Features

------

Ok, this kinda worries me....

From words of one of the engineers of the company Of matrokh:My made this chip in order to show all.we are as before capable of developing the foremost solution.Another worker answered less seriously:Just odds of fun..It is obvious that soon so expensive a solution will not be able to influence the repartition/ conversion of market, but to here improve renome of the company Of matrokh it is capable.Even if the number of sold maps/charts/cards on the base Of paryuelia-shchy2 does not exceed 10000.One only announcement of the chip of that containing the foremost (not having thus far analogs) technological solutions is capable to draw close attention to itself. Thus, in comparison with the foregoing generation of accelerators, the spacing strip of local video memory twice will increase.But speed is the not main horse of the products from Matrokh.To the surprise, the chip Of paryuelia-shchy2 is not equipped with any technologies of the savings of the spacing strip of memory.this fact is surprising.similar technologies are present in all last products from ATI and NVIDIA, but apparently Matrokh decided to accelerate and to reduce the cost of the development of the chip, which is for them faster symbol, than by the attempt to conquer market. The tests of engineering samples on the also very damp/raw drivers show only 20..30% advantage Of paryuelia-shchy2 over NVIDIA To geForche of 4 Tiya'00.Even under the condition of successful tyuninga of software, two-fold superiority will hardly become possible.One-and-a-half times.here is the upper limit.It is obvious that with the essential cost/value of maps/charts/cards on the base of the new GPU from Matrokh, to usual users has the sense to buy them for the quality and the unique possibilities, but not speed.From other side, Matrokh it cannot pretend to the success on the professional/occupational market without certified OpenGL of drivers (possibly, they they will appear together with the professional/occupational rule of maps/ charts/cards on the base Of paryuelia-shchy2 release/issue of which it is planned/glided).The map/chart/card is unjustifiably expensive for niche prosto of the amateurs of the qualitative of 2D..Thus, to the first period remain enthusiasts, poluprofessionaly, fanaty of stamp and we.specialists in videokartam: -).the possibility Of direchtKh 9.

[Geeforcer]

I must say I found the following slide strange:

• Some more info:
• All cards will have 256bit bus.
• 256MB cards will start around $500, 128MB cards - $400.
• All cards will be made by Matrox.
  
  Specks:
• 80M ts
• .15 um
• clock up to 350MHz
• 256 bit bus
• 64/128/256MB RAM
• 4 pipes
• 4 TMUs/Pipe (Why???- Geeforcer)
• Fillrate up to 1.4Gp/5.6Gt per second
• PS 1.3; 4 texture and 5 combiner stages per pipe, ability to "unite pipelines in pares"
• VS 2.0, 4 units
• EMBM/DOT3
• Fixed DX8 T&L with expended matrix blending and skinning capabilities
• 10-bit per channel
• 2x 400MHx, 10bit per channel RAMDACs
• Full 10bit gamma correction table
• DVD and HDTV - 10bit output
• 2 Independent CRT outputs
• Adaptive SS FSAA up to 16x, only affects polygon edges.
• DX9 N-patches with adaptive tessellation
• Displacement mapping
• Glyph Antialiasing - edge smoothing and gamma correction for fonts

[MfA]

I hope this does not indicate curved surfaces in DX9 will be an extension of N-patches :( The hardware has to treat them as polynomial surfaces anyway, that should be exposed to the developer. We cant always keep dicking around with low poly modelling, and thats the only thing N-patches are good for, existing tools for high detail modelling with curved surfaces use the classic polynomial surfaces for good reason.

[Geeforcer]

I could translate the whole article, but that would take too much time. However, if anyone wants any part of the article properly translated (as opposed to crappy machine translation), tell me and I will do it for you.