NV30 Update

[CMKRNL]

Some new information. Not from someone on the project, but close enough.

---------------

-NV30 just completed INITIAL tape-out (this week)
-First OEM samples end of October
-Foundry is TSMC, not UMC
-There were several things that had to be dropped in the design when they realized they were going to be very late with the part. This decision was made several months ago. The most notable feature was the primitive processor. This will show up in NV35.

----------------

What surprises me here is that they actually think that their part will be bug-free enough to ship customer samples from the initial batch! They must be supremely confident in TSMC's 0.13u process as well as their own design. More than one metal rev and they're going to blow December production as well.

 

[Sabastian]

What??!! Where do you get this info from? If in fact the initial tape out has JUST occured there is no way the nv30 can be shipping this year. I may be wrong but I have serious doubts about this claim you are making. Nvidia will have to renig on its nv30 launch time frame .....

 

[MDolenc]

Guys why don't you check this link a little?

 

[Joe DeFuria]

Guys why don't you check this link a little?

What is there of relevance? (Am I just missing it?)

Is it about the contest in which one of the prizes is NV30 based systems and hardware? They state this in terms of the prizes:

Please note that for the above prizes, we can not promise a shipping date. The prizes will be sent out when they're available, but the exact shipping dates (particularly for prizes in the NV30 category) have not yet been revealed.

 

[Dave Baumann]

the question is whether the initial 120M number of transistors quoted includes the things that were in the design but are now supposedly dropped...

 

[Joe DeFuria]

Well, in any case, I hope we can finally put to rest nVidia's mythical "flawless execution" claims. Gee, with this, plus the latest buggy Detonator driver release, you'd think nVidia was "just another chip company". (Which of course, they are.)

It's amazing what direct competition does to expose "reality".

 

[Richthofen]

you seem to forget that the latest detonator driver is a beta version.
Compared to the other company which sells boards and drivers which are nothing more than a huge bug, this is nothing.

 

[Sabastian]

the question is whether the initial 120M number of transistors quoted includes the things that were in the design but are now supposedly dropped...

Yeah but that won't matter too much as long as it is DX9 compliant and faster then the Radeon 9700..... interesting development. Nvidia may go for the no frills move once again in the name of expediency. I wonder just how much was dropped outside of this "primitive processor" they talk about? I wouldn't surprised if it comes in at a lower transistor count then the Radeon 9700 as well. But doesn't DX9 compliancy imply a larger transistor count? I mean that features of DX9 require more hardware don't they?

 

[Bjorn]

the question is whether the initial 120M number of transistors quoted includes the things that were in the design but are now supposedly dropped...

That's something that will be really interesting to see. Especially after all the claims regarding the impossibility to make the NV30 on a 0.15 micron process since we all know that it's very possible to make a 110 million transistor part on 0.15

 

[MikeC]

Gee, with this, plus the latest buggy Detonator driver release

I'm sure you're aware that they are beta drivers.

 

[Joe DeFuria]

Um, yes, I'm aware they are "beta" drivers. However, they are officially released betas, not some "leaked" driver. They are also for an EXTREMELY mature architecture, the GeForce3/4.

And I'm sure you are all aware that the 9700 drivers are "version 1" for a brand new architecture.

The point is, I don't see any difference in "driver quality" for ATIs and nVidia's products. And truthfully, I haven't seen much of a difference since the GeForce3 / Radeon 8500 era...when you compare driver state relative to the introduction of the new architeture.

It other words, when NV30 is finally released, I suspect the NV30 drivers will appear "much more buggy" than the R300 drivers at that point in time. ATI's driver team will have had quite a few months to work out the initial bugs, and developers will have had some time with the board as well.

 

[Sabastian]

the question is whether the initial 120M number of transistors quoted includes the things that were in the design but are now supposedly dropped...

That's something that will be really interesting to see. Especially after all the claims regarding the impossibility to make the NV30 on a 0.15 micron process since we all know that it's very possible to make a 110 million transistor part on 0.15

Hrm.. we know that ATI can do it that is for sure. But to take a design from .13 micron to .15 micron would require some effort. I am not so sure that nvidia will even bother doing such a thing. For some reason the nv30 design was created with the .13 micron process. Now why would you suppose that Nvidia decided that it should be this way? My first conclusion is that nvidia thought that the NV30 architecture would be too hot for the .15 micron process. Given their track record with the nv25 needing so much cooling with about half of the transistor count. My guess the nv30 at the .15micron process would be too hot. JMO of course.

 

[Dave Baumann]

That's something that will be really interesting to see. Especially after all the claims regarding the impossibility to make the NV30 on a 0.15 micron process since we all know that it's very possible to make a 110 million transistor part on 0.15

Well, I would highly doubt they would be moving back to .15um. They need to go through the .13um barrier at some point so why not with a part thats already late? R300 was supposedly 'hand tweaked' to reach the speeds that it does, which takes a lot of time on a desing like this, I'd have my doubts NV would be in a position to be able to spend that time on NV30 with .15um - probably better spent now just getting over the .13um hump.

However, to me its more interesting in relation to the recent rumours that NV30 spec was upped in the face of R300 - if the 120 number was from the design of old but it still comes out to be about 120M then that could give some credence to that.

However, in reality we will never know this and its all based on speculation. Really, this thread is nothing more than more speculation either.

 

[Joe DeFuria]

I don't think Bjorn is suggesting they redesigned for 0.15.

It's just that it will be interesting if the NV30 turns out to be "only" say, 90-110 million transistors after removing some functionality. Then we'd have nVidia on 0.13 with less transistors than ATI at 0.15.

 

[Doomtrooper]

you seem to forget that the latest detonator driver is a beta version.
Compared to the other company which sells boards and drivers which are nothing more than a huge bug, this is nothing.

You know I have a problem with that, the main reason being the Geforce 4 is basically the same architecture as the Geforce 3..they use a UDA architecture yet after almost two years of working with that design, how can they release a BETA driver

 

[duncan36]

Fri 7:36am NVDA InPlay: NVIDIA may delay new chip - Briefing.com

Well whoevers paying for this to show up on the Nvidia newsfeeds is fairly certain that the Nv30 is being delayed until next year.

 

[Bjorn]

I don't think Bjorn is suggesting they redesigned for 0.15.

It's just that it will be interesting if the NV30 turns out to be "only" say, 90-110 million transistors after removing some functionality. Then we'd have nVidia on 0.13 with less transistors than ATI at 0.15.

Exactly. As i said, Nvidia has again and again said that the NV30 needs a 0.13 micron process because of it's advanced funtionality and what if it turns out to be a <= 110 millon part ? (since we already have a 0.15 micron 110+ million DX9 part out on the market)

This of course depends on what you define as "needs"

I have to say that i don't really care that much about marketing crap like that but it's always fun to see companies PR stunts go wrong. (First DX 9 part anyone ?)

 

[Sabastian]

Exactly. Nvidia has again and again said that the NV30 need a 0.13 micron process because of it's advanced funtionality and what it it turns out to be a <= 110 millon part ? (since we already have a 0.15 micron 110+ million part out on the market)

Again I think that you are assuming that nvidia using their NV30 architecture can do this. I think that their technology is hot running by its nature.(assuming they are using the same tech found in earlier chips.) Nvidia chips have been running hoter then ATIs even at relatively the same transistor count with lower clock speeds(on nvidia tech) at the same .15micron process. (eg. Radeon 8500 vs geforce 4 ti series.)

Even though the Radeon 8500 carried only in the range of 3 million less transistors then the Geforce 4 ti series cards the Radeon 8500 was considerably cooler running even with higher default clock rates.

 

[Bjorn]

Again I think that you are assuming that nvidia using their NV30 architecture can do this...

Nope, i'm not assuming anything. What i'm saying is that Nvidia tell's the world that they need 0.13 micron process because the NV30 is so advanced. And, that that will look a bit bad with regards to the competition. That is, if the NV30 turns out to have less amount transistors then the competing card on a 0.15 micron process.

And what do you mean by "using their NV30 architecture can do this" ?

 

[Galilee]

It other words, when NV30 is finally released, I suspect the NV30 drivers will appear "much more buggy" than the R300 drivers at that point in time. ATI's driver team will have had quite a few months to work out the initial bugs, and developers will have had some time with the board as well.

Why would the NV30 drivers be more buggy than R300 drivers? That seems strange. But sure they might be more buggy than R300 drivers in five months.

And plz, "version 1"? hehe, Just like Detonator drivers they have been around since Radeon 1.