Nv40 16 full pipelines- The Inq.

I dont think the nV40 has 16 pipelines. I'll be amazed if they even have 8x2 with shaders 3.0. As for ATI I would not be amazed if they come up with an 8x2 design right now, something like a 16x1 is quite possible though, considering what they did on .15 microns.

100+ million transistors at 412mhz, on .15 microns? Quite amazing and the chip still has room for speed. People tend to forget that the biggest chip nVidia did on .15 was the good old nV25, geforce 4 ti family, at "only" 300 MHZ, and that their nV3x family has always being in a more advanced tech, and not able to compete anyway.

All I am saying is that Ati did amazing engineering, and that they have had 1.5 years working on the new chip. But, I will say just wait and see, since nVidia must have done everything to catch up all this time.

 

Fair questions and much better than nVidia cheap-shots. Deserved or not deserved; what's the point? What if questions are wonderful for speculation. I remember when many didn't want to entertain the idea that the GeForce 2 would run at a core clocking of 200 and be 4x2.

Probably what the GeForce 2 did for nVidia if the rumor is true. Consumers will forgive and forget if they know the product delivers the goods, imho. There is a lot on the line for nVidia to regain their proud name recognition. nVidia needs to make a statement, imho.

That was exactly the reasoning behind why many didn't think the GeForce 2 would be 4x2. I don't think the product will have enough raw bandwith like the GeForce 2 -- if the rumor is correct. However, that didn't stop the GeForce 2; and that 1600 number was a powerful selling point. I wouldn't be suprised if the rumor is true; I wouldn't be surprised if the rumor isn't true.

 

Holy crap, page 22 of 21. Maybe this will bump it.

 

I think that may have been the worst analogy I've ever read... .

 

hehe I was wondering why there were posts that I knew were there but I couldn't see...something is definately a little funky with this board.

 

Its a damn glitch(on the board). LOL

 

My own guess is that as soon as the R300 came out, Nvidia took a look at NV30 and then sent out an ukaze to their other design team: NV40 must clobber the 9700 Pro to at least the same degree that it beat the Ti4600. NV30 was too far along for Nvidia to react to R300, but NVV40 wasn't.

Nvidia does have good engineers, so as long as NV40 isn't hobbled by non-DX9/OGL design constraints I think they'll succeed at that goal. However, that may come at a serious cost in heat, power, yields and therefore margins. ATI is on record as having said they are willing to accept lower yields and margins on R4xx in order to hold the performance crown, so Nvidia doesn't have a choice but to go all out.

The question is will Nvidia be able to do it, and if so what architecture will they be able to pull off to do it. Past evidence has indicated they're more than willing to play fast and loose with the definition of pipeline so any claims along that line are subject to scrutiny. It may be that the NV4x architecture can't be described in terms of pipes-- we don't know. The only thing we do know is that Nvidia has a hard sell ahead of them because not only do they have to come up with a very fast part, but they also have to overcome all the questionable optimizations and outright cheats of the past generation.

It won't be easy, and unless they can do it without resorting to the questionable tactics they've been using it may be a hollow victory. There are people who will avoid their products purely because of the tactics, and it will take a while for them to build a new reputation-- good or bad.

 

NVIDIA told FiringSquad that 44.12 or something was the first driver that had a shader compiler.

 

Whether it was the first compiler or not would depend upon what you call a compiler.

If you take the definition that software which directly translates one instruction set to another, with no processing dedicated to reordering instructions or otherwise modifying the algorithm, is not a compiler, but simply a translation layer, then there was a specific driver set where the compiler was first implemented.

 

uh... This is Total BS. And likely BS from Nvidia themselves. They dont even believe in "pipelines" anymore remember? Their Architecture is completely Different than what most people consider a Traditional Pipeline layout.

The Nv40 is a 16 *PIXEL* Architecture in the exact same way their cuurent lineup is a 8 PIXEL architecture. And then its not 8 or 16 all the time.

 

You were on vacation HB? 8)

 

Something like that

 

Hehe good to hear

 

Well, to be more clear: an optimizing compiler for shader code then.

Obviously a compiler is anything that takes code and produces a different target code. Optimization is only one (optional) step. I don't think it's too presumptuous to assume that when most people talk about "The Compiler" in NVIDIA's drivers they are in fact talking about the shader compiler which produces optimized code. After all, that is what has been publicized.

 

I presume he's seen what I have seen, and that was my reaction too.

 

Can you tease us more precisely ?

 

If it is the NV45 that's gonna be released next as the NV40 when can we expect it? June/july/august? instead of April/May?

When are we gonna see the R420 out? If ATI wait for Nvidia to release and by june Nvidia might have the NV45 out on silicon by then.

I wouldn't mind ATi releasing the R420 now on us plebs atm.

I suggest that ATI get the R500 chunking NOW! (although I suppose they already are)

US

 

Yeah, that'll be interesting (both in terms of how they've implemented it, how it performs and how they market it).

There sure was a lot of "ooh-ing" and "aah-ing" after people went under NV40 NDA.

MuFu.

 

NV40 is fucking impressive.

(And no, Dave didn't tell me its specs even though he knows under NDA! Hey... there are some insiders that still likes me although I am no longer that interested in knowing specs before they are public!

 

More "ooh-ing" and "aah-ing" than there was after people went under R420 NDA?