[pascal]

From Ace´s Hardware:

3DLabs: 76M Transistor Graphics Chip This Christmas (HARDWARE)
By Brian Neal
Tuesday, March 12, 2002 9:55 AM EST
As we reported yesterday, Creative plans to acquire 3DLabs with the intention of driving 3DLabs' 3D graphics technology into the mainstream desktop 3D market while continuing to support the professional 3D space currently occupied by 3DLabs. This re-entrance into the mainstream desktop 3D market will occur with a new graphics architecture 3DLabs has been developing which is said to be highly scalable and programmable. After listening to Creative's webcast and Creative CEO Sim Wong Hoo's comments during the Q & A session, we have learned some new details about this new graphics architecture.

The new architecture has been described as being "very different" and "revolutionary" in comparison to other products currently on the market. It is said to have very high quality output and accuracy, which we speculate could potentially mean support for 64-bit color. Mr. Sim Wong Hoo also described the technology as "screamingly fast" and declared it the world's first single-chip tera-op graphics processor with general purpose programmability. The chip is rated at 170 GFLOPS and consists of 76 million transistors. In response to a question asking about the time-to-market for this new chip to reach the mass-market, Creative replied by saying, "We fully expect to be in the retail market by Christmas of this year."

Creative also talked about the business side of the graphics market, indicating that the product cycles for sound products are generally 2.5 to 3 years long, while acknowledging that product cycles in the desktop 3D market are much, much shorter. In response to a question regarding the possibility of a sound and video combo product being produced, Mr. Sim Wong Hoo stated that the company has dabbled with combo products in the past with little success because the result of combining separate products/technologies with different life cycles is usually the worst of both worlds.

3DLabs will operate as a separate entity and will also sell graphics chips to other manufacturers. It is not entirely clear to us whether or not this includes the new scalable graphics chips targetting the mainstream market, however, because the webcast was difficult to hear at times. Overall, it looks as though this merger could be very promising. Despite the lack of specific details regarding 3DLabs' new architecture, the fact remains that we now know there will be a new 76 million transistor 3D graphics solution in the mass-market by the end of the year, whereas two days ago we saw very little serious competition at all for ATI and NVIDIA, particularly with the uncertainty surrounding the STMicro sale. We will continue to report on this new 3D graphics architecture from 3DLabs as we learn more.

Link: http://www.aceshardware.com/#55000461

My guess this chip (which probably has a working prototype) was enough to make Creative buy 3DLabs.

The fact that it is programmable chips will have no immediate impact for any gamer (based on previous DX8 experience) until maybe 2004.

What I am really interrested in is how well will it plays next generation games (DX8 and OpenGL):

- This "very high quality output and accuracy" is wellcome and maybe it is a combination of 64bits and the traditional professional quality of 3DLabs cards (see it http://www.3dlabs.com/product/techno...at_quality.htm ). Something like Supersene will be great 8) http://www.3dlabs.com/product/techno...tialiasing.htm. I just upgraded from a Radeon 32MB DDR to a GF3 Ti200 and I must say that my Layman´s eye dont like much the GF3 image quality .

If this new chip is reasonably fast and have a great image quality then I am in. I doubt ATI and specially Nvidia will have this quality.

Games like Doom3 with a BIG focus on great lighting using multiple pass will require high quality to be fully enjoyed. 8)

[Nappe1]

This will be a wild year to the buyers.
- Matrox
- ATI
- nVidia
- Creative
have been already signed to race and more pontential racers are:
- Bitboys with their Next gen
- S3/VIA with their Columbia
- PowerVR/VIA/STM/IMG (or whatever that is...) with their STG5000 series.

Me more than likes!

What more we would like to hear?? maybe Micron announcing that Rendition is coming back?

[RussSchultz]

Originally Posted by Nappe1

This will be a wild year to the buyers.
- Matrox
- ATI
- nVidia
- Creative
have been already signed to race and more pontential racers are:
- Bitboys with their Next gen
- S3/VIA with their Columbia
- PowerVR/VIA/STM/IMG (or whatever that is...) with their STG5000 series.

Me more than likes!

What more we would like to hear?? maybe Micron announcing that Rendition is coming back?

You are ever the optimist.

Me, the pessimist, thinks the only people who will release anything this year are NVIDIA and ATI. Matrox might, the rest are just pipe dreams.

[Joe DeFuria]

4Q 2002 / early 2003 should be extremely interesting indeed. I would guess that Matrox will be in the battle. I think 3D Labs will eventually make it into the battle, though I would be surprsied to actually see the product on the shelves this year. Projected dates as far as x-mas tend to slip...

I'll bet Carmack is getting blitzed with "support" from all of these graphics companies. X-Mas 2002 / early 2003 sounds like it should be just about the right time to see Doom3, if not the Doom3 "test". Whichever of these next-gen parts benches the best on that engine, is going to get a huge boost...no matter how poorly that part might play anything else.

[pascal]

IIRC Carmack gave an interview to a magazine (gamesomething ??? ) and he said that the Doom3 engine will probably be finished by august/2002.

[ben6]

Micron/Rendition is all but dead in the graphics chip /chipset arena. They had a dual channel DDR solution ready, but Intel refused to grant a FSB license (similar situation to Nvidia) and bought the rights to their chipset. You might see parts of Micron technology in Intel chipsets though... Of course my buddy at Micron is now working on Smartflash cards or the like (he used to work for Diamond Multimedia)

[psurge]

Question, how far from completing a chip can you be when you already know the transistor count of your chip? (regarding 3dlabs 76million transistor announcement). Also, who fabs chips for 3dlabs?

[3dcgi]

Originally Posted by psurge

Question, how far from completing a chip can you be when you already know the transistor count of your chip? (regarding 3dlabs 76million transistor announcement). Also, who fabs chips for 3dlabs?

You can have a pretty good idea of transistor count fairly early in the design process so the number doesn't really tell us anything. However, if they are expecting to have products by Christmas they've probably already taped out the first revision of the chip.

[flask]

3dlabs completed tape-out last year,
but were looking for money backing,
which they have now...

[Simon F]

Originally Posted by psurge

Question, how far from completing a chip can you be when you already know the transistor count of your chip? (regarding 3dlabs 76million transistor announcement).

This is just idle speculation, but perhaps it might be determined by other factors...
Perhaps they decided what the chip should cost to manufacture. Cost is largely determined by the area of the chip (i.e. N*M millimetres) and for a given process (eg .1x u) that would determine how many gates/transistors they can use. <shrug>

[pascal]

Look at this:
- 170 Gigaflops (GF3 is only 70Gigaflops)
- 76 millions transistors
- by the end of the year
- mass market target

edited: it is 2.2Gigaflops per million transistors, maybe a high frequency clock.

Maybe it will be a high clock (300~400MHz) .13 micron process chip. It makes sense to me. 8)

edited: TSMC ???

[pascal]

More about Doom3 I got from U2 forum

Based on what information? All we know is Carmack said the next Nvidia and Radeon cards will be out before Doom 3 so it could be early 2003. We also know at quakecon Carmack said it will take him about a year to finish the engine. So if indeed the engine is done by August, Spring 2003 is still quite possible I doubt id even knows when its gonna be out if they are truly following there "When Its Done" philososphy.

Anyone here has been at this quakecon?

[hughJ]

the gamespot video interview was from quakecon..

I know this has been posted here before, and it's relatively old, but I guess I may as well post the link to what I was referring to :P

http://gamespot.com/gamespot/filters...469881,00.html

"...I've still got at least another year of making things look better and faster on there[the engine]..."

[pascal]

The latest 3DLabs cards have really impressive specs with 256bits DDR memory access:

Frame buffer: 128 bits
Texture buffer: 64 bits
DirectBurst: 64 bits

http://www.3dlabs.com/product/wildcatIII_6110_index.htm

What is DirectBurst ?
I would like to see some screenshots of Q3 or UT with Superscene. 8)

edited: Hughj, Great. I bet we will have at least a Doom3 test by Q3/Q4 this year 8)

[arjan de lumens]

"DirectBurst" appears to be a memory dedicated to holding onboard copies of (pre-T&L) vertex arrays, so that the board does not need to re-fetch vertices over the AGP bus for every time it does T&L on them.

[pascal]

Interresting, the new Wildcat III has a single chip doing the T&L and rasterizer functions (the Wildcat II has two chips).

It means a chip with 256 bits memory access not using UMA (Three address bus 28bits for texture, 27bits for framebuffer and 25bits for directburst, total 100bits for addressing driving a lot of chips !!!).

The next 3D Labs chip will need to be faster than willdcat III with Pro Applications. Will it use 256 bits too? The same way or different way?

[3dcgi]

[quote="Simon F"]

This is just idle speculation, but perhaps it might be determined by other factors...
Perhaps they decided what the chip should cost to manufacture. Cost is largely determined by the area of the chip (i.e. N*M millimetres) and for a given process (eg .1x u) that would determine how many gates/transistors they can use. <shrug>

[Simon F]

Gosh.. still patiently waiting for 3dcgi's comment

[pascal]

Correction, the total address bus for Wildcat III is 80 bits

Anyway this non-UMA schema exploit a lot the eficiency of the Locality Principle.

Looks like the direct burst use only SDR SDRAM.
Maybe a more eficient way is dual 128bits DDR external memory, one for framebuffer and the other for the rest. It means more bandwith for textures (when needed) and less addressing bits.

[Geeforcer]

Originally Posted by Nappe1

This will be a wild year to the buyers.
- Matrox
- ATI
- nVidia
- Creative
have been already signed to race and more pontential racers are:
- Bitboys with their Next gen
- S3/VIA with their Columbia
- PowerVR/VIA/STM/IMG (or whatever that is...) with their STG5000 series.

Me more than likes!

What more we would like to hear?? maybe Micron announcing that Rendition is coming back?

That what some people said in the beginning of last year about 2001.

[3dcgi]

Originally Posted by Simon F

Gosh.. still patiently waiting for 3dcgi's comment

How did that happen? I guess I didn't read my post after hitting submit. Anyway, I was just saying that you are right in that cost does affect size from the beginning, but 76 million transistors is awful specific to be an early estimate.

[LittlePenny]

Originally Posted by 3dcgi

but 76 million transistors is awful specific to be an early estimate.

I honestly know little about electrical engineering, but when referring to 76 million as a specific transistor count just awes me, that a number rounded in the millions is specific.