Nv40 16 full pipelines- The Inq.

hey this thread made 32 pages. (now maybe 33)


NV40 = 32x0 ?

 

actually, GF1 had two TMUs originally. one didn't work correctly, so they cannibalized it to form a single trilinear TMU. GF2, though, has two bilinear TMUs.

 

That's the first time I've ever heard that one. Where do you get it from? Since the GeForce2 didn't have much performance difference between trilinear filtering and bilinear filtering, I have to assume that the GeForce 256 also would not have had much performance difference, so I see little reason to have that second TMU, except for PR reasons.

 

That's my sole flaw.

 

Conversation with Uttar. "NV15 is a 4x2, but each TMU is bilinear-capable. NV10 is a 4x1, but trilinear is free (fillrate-wise, at least). It was actually a bug in the NV10 design, which they fixed in the NV15, and then they boasted about how great their doubled multitexturing fillrate was. It [the NV10] was supposed to be 4x2 bilinear, but it didn't work somehow. So, they made it 4x1 trilinear."

Of course, he told me he got that from other people, so somebody will confirm or deny soon enough.

 

More than 1 year. In august it will be 2 . SO a year and a half .

If the r420 is better than the nv40 then it will easily be 2 years .

 

That's what I said a little while back. TheInq is da Stink.

US

 

I read the Inq because it's entertaining. It may not be news, but it's fun.

 

Uttar has been known to be wrong in the past.

 

can't believe they are actually going for it

 

And how do you supose GF1 had free trilinear? It takes 2 (two) texture accesses. Or maybe they designed it that way because we all know how important it is for them to have all textures on screen with full trilinear.

 

That's an exaggeration. The Inq might be wrong a lot, but they're right a lot too. How quickly one forgets that they uncovered the hidden Intel X86-64 plans.

 

It should be obvious that I used the tilde sign on purpose, indicating "more or less" for both cases. There's no severe error or exaggeration in the rounded up numbers and I didn't of course address future posibilities either.

 

okay i thought u meant it as less because acording to my memory nvidia held hte market for less than 3 years.

 

Oh, I'm sure *that* was a challenge.

Come on. Anybody who has an ounce of intelligence could have made that rumor up. If you make a lot of guesses, you're bound to be right every once in a while.

 

Heh, it was people just like you who were poo-pooing the Inq Intel 64-bit extensions stories.

Of course, assuming NV40 launches with 16 full pipelines and 210m transistors you'll be too arrogant to admit you are wrong still and say that was a guess too I suppose.

 

Arrogance? I just don't like websites posting rumor as truth. Anyway, I never "poo pooed" Intel coming out with an x86-64 architecture. It seemed rather likely, given that they were going to have to do it sooner or later, and that it was unlikely to get IA-64 to work in the desktop within that time (64-bit processing is required for memory sizes above 4GB, and the desktop will be above that quite soon).

 

Slightly OT, but how do you feel about it when websites post up rumors as rumors? (I tend to do that a lot so I thought I'd best check. )

 

Chalnoth: Sorry for the french source, tons of english sites say it too, just couldn't find any whose reliability is generally accepted, beside this one.

Rough translation: The GeForce 256 is thus capable of rendering 4 pixels in single texturing or 2 pixels en dual texturing ou 1 pixel en quad texturing per clockcycle, no matter whether it's in bilinear or trilinear filtering.And AFAIK, this was not the case for the NV15. NVIDIA's marketing department was smart enough, though, not to make notice of this to most review sites at the time.

Uttar

EDIT: As for the REASONS of this choice, a few possibilities:
1) It's a bug (rather big one if you ask me, eh - so kinda unlikely, but possible). Two bilinear TMUs, but they lack the functionality to be able to
2) It's intentional; the chip is already bandwidth starved, after all! Trilinear increases the caches' efficiency. And they might save a few transistors this way; probably not many though.
3) It's a marketing choice: they didn't need 4x2 against 3DFX for this generation, so pushing trilinear first and then pushing dual-texturing once the GF2 came out makes sense.
Or better, a combination of 2) and 3)

 

I was under the impression that The Inq did just that... I never noticed that they were taking themselves too seriously...