Will NV45 use TSMC's 90 nm ?

[Mod]

"In addition, Nvidia will use the more advanced 90nm process once TSMC is ready, Huang added."

I guess this technology will be ready by the end of this year. So, it will be ready to be taped out for NV450, released around september-october of 2004 ( if the semester cycle keeps respected).

(WOW ! Doc told me that from 2015 :D )


http://digitimes.com/NewsShow/Article.asp?datePublish=2002/12/17&pages=02&seq=6

[gokickrocks]

i dont think we will see it in 04, in another digitimes article nvidia said the product cycles are more likely going to be lengthened

[Philibob]

If not 2004 then it should be the first cards in 2005. CPUs using 90nm should be out next year so if a design of a card starts then and that takes 2 years that'll be 2005

[borntosoul]

the whole nv3X design is based on having a top to bottom product line based on that design , so that basic design will have a long life cycle but a different road will be taken on the high end in future ,that will be as needed to keep technology leadership and test for new tech that will trickle down to the mid low end later

[StealthHawk]

Originally posted by gokickrocks
i dont think we will see it in 04, in another digitimes article nvidia said the product cycles are more likely going to be lengthened

why would they lengthen their product cycles when ATI is accelerating theirs? that makes absolutely no cents(pun intended)

[gokickrocks]

you make it sound as though i could read the minds of the engineers at nvidia...im just going by what mr.huang or whatever said

[StealthHawk]

Originally posted by gokickrocks
you make it sound as though i could read the minds of the engineers at nvidia...im just going by what mr.huang or whatever said
ok then, where did he say that. link please.

[Uttar]

nVidia said their NV40 will be made on 0.09 Micron
Which is, if you didn't quite get it, 90nm :)

And for the billionth time, damnit, nVidia is on a 18 month cycle.
One question remains... Will they consider the NV30 delay as delaying every of their future generation products? Or, since it taped out in August, were they free to already begin the NV40 design in a timely schedule to be ready in H1 2004?

IMHO, the NV40 will either come in late H1 2004 or early H2 2004


Uttar

[gokickrocks]

http://www.digitimes.com/NewsShow/Article.asp?datePublish=2002/12/17&pages=02&seq=3

last sentence...

"Meanwhile, with increasingly complex product design and more advanced process technology, Huang predicted that the interval between product generations will lengthen."

[Mod]

It makes sense. The number of transistors doubles each 12 months for videocards, whereas the size of the transistor decreases ~40% each 18 months ( that is, you can double the density every 18 months).
So, from now on, video cards will have to respect moore's law, doubling the number of transistors every 18 months instead of 12 months. Otherwise, the chips will get enourmous.

[nutball]

It also makes sense given the sheer complexity of the newest generation of GPUs, that the product cycle will start to lengthen.

They now far outweigh a CPU in terms of transistor count in the core (ie. not counting the on-die caches of CPUs).

Just verifying the design must be a real headache.

I dare say ATi's cycle will slow too, though doubtless they'll come over all macho and deny it.

[StealthHawk]

Originally posted by nutball
I dare say ATi's cycle will slow too, though doubtless they'll come over all macho and deny it.

everyone keeps saying that R400 will come out in Fall of 03, meaning that they're staying with their 1 year cycle of new architectures.

the key word mentioned to me was "product generations." now does this just refer to new architectures or does it include refreshes too?

if it includes refreshes nvidia is obviously in a lot of trouble as ATI seems to be escalating its product cycle to 6 months to include a major refresh.

[Bigus Dickus]

Originally posted by Uttar
And for the billionth time, damnit, nVidia is on a 18 month cycle.

Saying that repeatedly doesn't make it true. nVidia has been on an 11 month core release cycle (NV15, NV20, NV25, NV30), 22 month new architecture (NV20, NV30) for over two years. Long enough to call it a trend at least.

[nutball]

Originally posted by StealthHawk
everyone keeps saying that R400 will come out in Fall of 03, meaning that they're staying with their 1 year cycle of new architectures.

Well the way I read the comment from the NVIDIA CEO was more a "over the next few years you're going to see this", rather than "after NV30 we're going to a 24-month cycle".

OK, so R400 turns up in the autumn (maybe). What about R500, R600, etc., etc., etc?? Quite frankly, if those are coming out on a 12-month cycle, then they're either going to be a) evolutionary, not revolutionary, or b) be half-arsed and not work properly. I really don't believe the pace of development in the GPU area can accelerate much further!


the key word mentioned to me was "product generations." now does this just refer to new architectures or does it include refreshes too?


Who knows. Maybe we'll start to see more refreshes of an underlying architecture in the future.

[Uttar]

Originally posted by Bigus Dickus
Saying that repeatedly doesn't make it true. nVidia has been on an 11 month core release cycle (NV15, NV20, NV25, NV30), 22 month new architecture (NV20, NV30) for over two years. Long enough to call it a trend at least.

Eh, yeah, saying something doesn't mean it's true :)

However, nVidia really wants to be on a 18 month cycle for new cores ( NV1x, NV2x, NV3x, ... ) .
They just aren't too good at making that happen :p

nVidia publicly aknoledged their goal for the NV30 was having it ready in August 2002 and everyone know the NV20 was expected to release in H2 2000 - nVidia just had a 6 month delay as with the NV30.

Problem is, even if nVidia is quite good at making sure they get a new product in stores every 6 months ( for H2 2002, that was the nForce 2 & the NV18/NV28 ) , they hate doing compromises. I guess management doesn't care, but engineers simply seem refuse it.

David Kirk agreed to say ( at the extremetech article I think, not sure ) that there were two reasons for the NV30 delays: 0.13 *and* 128BPP everywhere.

Now, that proofs part of the delay is because nVidia engineers refused to compromise. Bad engineers! Bad!
I'd guess that since they already did it twice, management would simply kill them if it happened a third time with the NV40. And when I say "kill"... I mean it! :)

Of course, only time will tell that.


Uttar

[Mod]

If you take only the chip size, within 5 or 6 years, you will have some that are the size of mother boards, if the number of transistors keep increasing faster than the tax of miniaturization of transistors.

I think delaying the cycle is more a problem of physics,than engineering or economics.

[Uttar]

In case anyone doesn't know it yet, just figured I would say it anyway...

0.13 ( NV30 ) to 0.09 ( NV40 ) is a bigger decrease than even 0.18 ( GF2 GTS->GF2 Ultra, not GF2 Ti ) to 0.13 ( NV30 ), and a much bigger one than 0.15 ( NV2x & GF2 Ti ) to 0.13 ( NV30 )

0.18x0.18 = 0.0324
0.15x0.15 = 0.0225
0.13x0.13 = 0.0169
0.09x0.09 = 0.0081

0.15 is 69% of 0.18
0.13 is 75% of 0.15
0.13 is 52% of 0.18
0.09 is 48% of 0.13
0.09 is 36% of 0.15
0.09 is 25% of 0.18

As you can see from these numbers, 0.13 to 0.09 is a bigger size decrease than 0.18 to 0.13! That means, from NV30 to NV40, we'll get as much of a transistor size decrease as from the GeForce 2 GTS to the GeForce FX!

Now, we've also got to remember 0.13 symbolized the transistion to copper. And that made the 0.15 to 0.13 transition very interesting in term of frequencies increases.

All we'd need to make 2004 a really, really exciting year is Black Diamond coupled with 0.09
There'd be only one way to summarize that: Yummy! :)


Uttar

[ZoinKs!]

I remember reading someplace that the next step for graphics cards would be .11, not .09

I don't remember where I saw that so it's definately rumor mill material for now...

[Uttar]

Originally posted by ZoinKs!
I remember reading someplace that the next step for graphics cards would be .11, not .09

Plans change quite often in the graphics cards industry :)


Uttar