[Typedef Enum]

I don't have the magazine, but one of the guys @ 3dgpu said that the latest issue of Wired magazine features an interview with nVidia's CEO, and he states that NV30 has, indeed, taped out and will contain approx. 120 Million transistors.

http://www.3dgpu.com/yabb_se/index.p...y;threadid=525

Taking this info @ face-value...I think it pretty much lays to rest the notion of a second chip...With a chip of that size, it seems very difficult to imagine a co-processor consuming even more real estate.

Maybe all that additional logic is for that Glide compatability we keep hearing about...

[BenM]

Holy crap! I have nothing useful to post, other than it seems like a monster.

[Pete]

Wow. So, are we guessing dual/quad pixel shaders? Perhaps more vertex shaders than R300's and Parhelia's four?

A doubling of NV25's die seems to be no small upgrade.

[Nappe1]

what tha hell they are building?? The Holy HandGrenade??
(check the Monty Python's movie 'Quest for holy grail.')

[mboeller]

How fast will it be?

120Mio Transistors @0,13µm <-> 63Mio @ 0,15µm.

Maybe it will only work at 250MHz. Or are most additional transistors for eDRAM?

[Typedef Enum]

How does that cliche go?

Expect the unexpected? That's what I'm thinking...Honestly, nVidia has had such a predictable history, it becomes really difficult to believe that they would veer off course, so to speak...But I think the realization that the competition was slowly catching up to her may have had some impact in this one.

I haven't read Wired to verify the authenticity of this claim...But assuming it's true, it just seems unthinkable that they could double the transistor count from a GeForce4...But if the CEO said it...it seems rather credible.

[Randell]

well they have hinted at a 'discontinuity' in their next chip havent they?

[PVR_Extremist]

So its gonna be a sound card

[ben6]

First announcement (remember I said sooner than later ) Now the wait for the 2nd shoe to drop on NV30 . Soon

[Evildeus]

Originally Posted by ben6

First announcement (remember I said sooner than later ) Now the wait for the 2nd shoe to drop on NV30 . Soon

You are the devil Ben6

This fall will really be interesting!

[Chalnoth]

Heh, Glide, riiiight....that'd be about 40 steps backward...and I think it would make everybody else in the industry very, very angry at nVidia (we really don't need another proprietary API). There is just absolutely no possibility that we'll see Glide ressurected.

But personally, I fully believe the 120 million transistor number. nVidia has been saying that this will be the most significant contribution to 3D graphics they've ever made. The big question, then, is why? Will the NV30 have much more flexible programmability (similar to the P10)? Or will it be more on the software side, with more accessible programmability?

Of course, we can be about 99.9999999% certain that the NV30 will have full 16-bits per color floating-point pipelines. But, this is really no more significant toward advancing 3D technology than 32-bit color was. And the usage of just more advanced pixel and vertex shaders wouldn't be any more significant than the initial introduction of pixel and vertex shaders.

So, why is this NV30 to be so significant? I'm really excited about finding out :)

[Gunhead]

Maybe something in the L part of TCL? 8)

[Nappe1]

Originally Posted by Chalnoth

...that this will be the most significant contribution to 3D graphics they've ever made.

I have no doubts about this, but I what I fear is huge amount of very advanced features that no one is willing to support yet, because it would be make game developing go back to direction where it was when proprietary APIs were only way to go. (regular version for "normal" cards and special version for NV30.) and I agree with Chalnoth that this is not a nowadays way to do this.

Cards that are too much ahead of their time, never gets fully supportted their feature sets at least with playable speeds.

for example, Pyramid3D was a way to much ahead of it's time. even if it would have stayed on schedule, It's features would have been fully supportted not until DX8 and for that it would have been a way too slow.
(think of it... card with equal memory bandwidth of TNT Vanta and with full DX8 feature set.... huh?)

another side of the coin is, somebody needs to take a step, eventhough it would mean feature set that gets no support at all during chip's lifetime.

[Dave Baumann]

And we don't necessaily know that, on chips these sizes, the supposed extra 13million transitors over the supposed 107 of R300 will actually necessarily bring much to the table in terms of features...

[Grall]

But think of it... What the heck could Nvidia use 120M transistors for?

eDRAM of some sort seems a bit unlikely in my opinion, they've never even hinted at being interested in that direction before. And 16-bit per color component FP pipes hardly consumes THAT much more either.

So are they simply heaping on (much) more of the same, like 8 pixel/vertex shaders etc etc? Or what?

I saw a list of possible topics for some Nvidia conference to be held around siggraph. I read stuff like solid modelling and non-graphics use of GPUs and stuff. Maybe that's what NV30 uses all those extra transistors for.

*G*

[Brimstone]

I read some time back that it will be possible for 2 AGP 8x slots to function on a motherboard. I have no idea if this is true, but if it is the case I could see NVIDIA allowing for SLI on their new cards.

Does anyone know if the AGP 8x spec allows for two AGP slots?

[pcchen]

AGP 8X (in AGP 3.0) supports a fan-out bridge for two AGP cards sharing a AGP bus. Of course, the motherboard manufacturer have to implement it.

[MfA]

Or maybe you can put it on the video card too? (Obviously it would make 2 chip setups a lot easier.)

[pcchen]

Ugh... I am not sure about this. IIRC there can be only one fan-out bridge on a AGP bus in AGP 3.0. If there is a fan-out bridge on a video card, what happens when it is put on a motherboard with a fan-out bridge?

Or maybe we need a sticker on the video card says "For MBs with one AGP slot only"

[pascal]

From TSMC process info NV30 with 120millions transistors the die area should be equal to a 90millions transistors .15 micron chip

This chip will not be cheap

If it has only 128bits bus then we can expect some big texture caches.

[Gollum]

Originally Posted by pascal

From TSMC process info NV30 with 120millions transistors the die area should be equal to a 90millions transistors .15 micron chip

This chip will not be cheap

If it has only 128bits bus then we can expect some big texture caches.

If a 90mil .15 micron chip is so bad, where does that put the 107mil transistor R300 at .15um considering cost?

[pascal]

I really dont know, but .15 micron now should have better yields than the .13 micron process (for now).

Also this 120millions come from the wired magazine nvidia CEO interview. Where did this 107 come from?

What I am asking myself is when we will have a sub $200 DX9 card?

[Nappe1]

Pascal: I guess that SiS will be showing Sub $200 DX9 card first...

In their roadmap it was marked to Q1/2003

[Ailuros]

Originally Posted by Nappe1

Pascal: I guess that SiS will be showing Sub $200 DX9 card first...

In their roadmap it was marked to Q1/2003

I have something far more capable in mind and with far more solid drivers

[Evildeus]

Some more info:
http://www.tweaktown.com/document.ph...Id=279&dPage=7

Derek Perez told us that we can look forward to film-like image quality with the NV30 and if this is true, there are certainly gonna be some exciting times ahead on the gaming front.