Fuad of The Inq, trying to explain NV40 ... AGAIN

[StealthHawk]

Yet we do not know whether or not the benchmarks in question were forced (via forceware hah) to run in brilinear -- but would NV really do that *if* NV40 has *that* much RAW power?

Why not? It just gives them that extra boost of speed. And most reviewers aren't smart enough to know the difference between trilinear and brilinear, even when looking at colored mipmaps

 

[digitalwanderer]

Yet we do not know whether or not the benchmarks in question were forced (via forceware hah) to run in brilinear -- but would NV really do that *if* NV40 has *that* much RAW power?

Why not? It just gives them that extra boost of speed. And most reviewers aren't smart enough to know the difference between trilinear and brilinear, even when looking at colored mipmaps

What's the big diff? Brilinear is almost the exact same thing anyways.

 

[StealthHawk]

What's the big diff? Brilinear is almost the exact same thing anyways.

That's what NVIDIA's been saying. So why backtrack? It just makes them look like idiots.

 

[volt]

What's the big diff? Brilinear is almost the exact same thing anyways.

That's what NVIDIA's been saying. So why backtrack? It just makes them look like idiots.

They are idiots for what they've done, but forget it I don't feel like speculating on filtering issue right now heh.

 

[duncan36]

Well the stock used to move on the Nv30 rumors and its not moving on these, if everything was running smoothly and the Nv40 was going to be on time and as dominant as people are saying the stock would be moving.

 

[ninelven]

Is that some sort of inverse / paradoxial logic?

 

[Miksu]

R420 is eight "extreme" pipelines, nobody knows what extreme means

Arh, this is getting really frustrating. First we hear some rumours about 12x1, then eight "extreme" pipelines is "confirmed" from multiple sources, then the 12x1-rumour hits us again and that also is "confirmed" from multiple sources. Don't tell me that we're back at the 8 extreme pipelines again.

 

[KimB]

Well the stock used to move on the Nv30 rumors and its not moving on these, if everything was running smoothly and the Nv40 was going to be on time and as dominant as people are saying the stock would be moving.

Do you really believe the shareholders have any better idea of what's going on? nVidia employees have already gotten investigated once on insider trading....do you think they'd risk it again? (side note: anybody remember what ever happened with that whole thing?)

No, all that you're seeing is that shareholders have become jaded since the release of the NV30.

 

[Ailuros]

What's the big diff? Brilinear is almost the exact same thing anyways.

That's what NVIDIA's been saying. So why backtrack? It just makes them look like idiots.

They are idiots for what they've done, but forget it I don't feel like speculating on filtering issue right now heh.

Advanced texture filtering is truly a complicated and large topic. Personally I don't think either/or IHV are immun to any possible criticism concerning that department.

Arh, this is getting really frustrating. First we hear some rumours about 12x1, then eight "extreme" pipelines is "confirmed" from multiple sources, then the 12x1-rumour hits us again and that also is "confirmed" from multiple sources. Don't tell me that we're back at the 8 extreme pipelines again.

Personally I'd prefer to see the 8 complex SIMDs with trilinear TMUs to be honest. Why?

Example:

12*500MHz = 6.0GPixels/s
8*500MHz = 4.0GPixels/s

For the first thinking of the usual bi-TMUs, once you think of trilinear filtering you drop to 3.0GPixels/s; having tri-TMUs in the latter case gives you 1.0GPixels/s more theoretical fill-rate.

 

[StealthHawk]

Advanced texture filtering is truly a complicated and large topic. Personally I don't think either/or IHV are immun to any possible criticism concerning that department.

Trilinear filtering is "advanced texture filtering?"

note: I'm not being sarcastic here.

 

[Ailuros]

I know you weren't sarcastic, but then again there's nothing wrong from what I can see in that sentence either, as long as "brilinear" could be seen in driver revisions on FXs even w/o AF enabled.

 

[volt]

It might have been considered advanced circa 5 years ago, but today it's considered a standard, not by NVIDIA of course.

 

[Tim Murray]

Trilinear TMUs are certainly not standard. That's why trilinear has a performance hit compared to bilinear on, hmm... all cards since GF256.

Will it have trilinear TMUs? As good a theory as any, but it still doesn't make a whole lot of sense.

 

[digitalwanderer]

What's the big diff? Brilinear is almost the exact same thing anyways.

For those of you that missed the "roll", I was being sarcastic....please quit PMing me about it.

 

[volt]

I'm talking about single pass trilinear which requires two or more TMUs -- there shouldn't be a big performance hit with that?

 

[vb]

I'm talking about single pass trilinear which requires two or more TMUs -- there shouldn't be a big performance hit with that?

Well, Nvidia considers trilinear on a x2 architecture a big performance hit, big enough to deny customers from using it.

 

[Hellbinder]

We understand that the NV40 will actually only have eight physical pipelines, but these will appear act like 16 in certain games and, indeed, in 3DMark 2001 Nvidia was telling people how 3DMark01 is a very nice benchmark since they can render 16 textures per pass in it, using only eight pipelines. What Nvidia is using is the ability of the Vertex Shader model 3.0 (known as PS 3.0 or VS 3.0) where the Shader can actually render textures as a virtual pipeline but can only render them without filtering information.

Link: Nvidia's NV40 mimics a 16x1-pipeline architecture with just eight pipelines

Essentially, nVidia is now abusing a 3-years-old fillrate benchmark? :? :?

incurable, incredibly confused.

thats Funny.. Now where have i heard that one before... oh yeah..

Its exactlly what i have been saying the Nv40 is for oh.. a couple months now.. an 8 pixel engine that can occasionally do 16 pixels under the right conditions.

Will be interesting to see what Nvidias "hardware support" for PS 3.0 is as well...

Likely not what some are thinking.

 

[Hellbinder]

R420 is eight "extreme" pipelines, nobody knows what extreme means

Arh, this is getting really frustrating. First we hear some rumours about 12x1, then eight "extreme" pipelines is "confirmed" from multiple sources, then the 12x1-rumour hits us again and that also is "confirmed" from multiple sources. Don't tell me that we're back at the 8 extreme pipelines again.

What we are at is that its important how many "PIXELS" in a given situation the hardware can produce. Which is not always the same. So 8 12 and 16 can all be accurate.

 

[digitalwanderer]

I KNEW someone was missing from the speculation-stew-special!


Hey Hellbinder <BIG HUGS!!!!>, what do ya know about the R420? I could use some fun musings, you think it's a 12 pipe? If so, think there is a chance at enabling more thru tweaking?

 

[MuFu]

AFAIK, the situation is...

8 "EXTREME" pipelines => what ATi told people to make them happy
12 pipelines => number enabled in R420
16 pipelines => total number present in R420

The main points of speculation now IMHO, should be...

~12 pipe by design or by default? Is the 4th quad pipe a "spare" or a "suprise"?
~whether the 8 "EXTREME" pipelines information actually has some basis in terms of architectural traits. Perhaps there is an 8 pixel/clk, FP32 mode?

MuFu.