eyebrow raising speculation on NV30-NV35

[megadrive0088]

someone posted this on Rage3D. I found it rather interesting. eyebrow raising, even. although this theory could easily get blown out of the water with just a few comments from people here who are 'in the know' ....of which, I am certainly not one.

My opinion on the situation is that the NV30 is the NV35.

Don't you think it's suspicious that nVidia says $400 million went into the production of the card, when the yearly R&D for both ATi and nVidia had previously been in the $100 to 150 million range?

I believe the rumours that nVidia had a 6 pipeline (probably with 2 TMUs ea.), .13 micron version of the NV30 taped out, and when it became apparent to nVidia that they were going in the wrong direction nVidia scrapped the NV30 and fast-tracked the NV35, re-badging it the NV30. When nVidia's CEO told Wired magazine around June 2002 that the NV30 had taped out, he wasn't lying... it's just that it was the 6 pipeline NV30 that had taped out.

That's why it's so late and so expensive; the $400 million would include the R&D of NV30 mark I and NV30 mark II (to use Ford GT40 nomenclature).

thoughts?

 

[Trawler]

must...resist...using...rolling eye smiley...

 

[MikeC]

My opinion on the situation is that the NV30 is the NV35.

Don't you think it's suspicious that nVidia says $400 million went into the production of the card, when the yearly R&D for both ATi and nVidia had previously been in the $100 to 150 million range?

NVIDIA's CEO stated that the entire NV30 product line will have cost them $400 million to develop. This includes the various models of the GeForce FX for the consumer desktop, the Quadro lineup (NV30GL), mobile, and integrated chipsets (nForce). Not sure about their plans for the Macintosh platform though.

 

[Laa-Yosh]

It has been explained several times that 6 pipelines are a bit un-practical, considering that computer design usually prefers numbers that are powers of 2.

Which reminds me about a question I had... lemme start a new thread.

 

[Sage]

umm..... yeah ok. im betting that $400 mil is figuring in some of the cost of bribes to make 3dfx sink so they could buy their tech

 

[KimB]

Sure, Sage, sure....

Regardless, the NV30 is a huge chip, and don't forget that nVidia has been getting samples back from TSMC on .13 micron since Februrary.

A good deal of that money was almost certainly spent attempting to get .13 micron to work right.

 

[megadrive0088]

so in all likelyhood, Nvidia should not have NEARLY the same problems with NV35, that it did with NV30. provided they don't have trouble with the NV35 itself. The .13 micron process problems are all ironed out now, or will be shortly. I don't see why Nvidia would not have a refreshed core with some enhancments + clock increase, by June-July. they almost have to with R350 coming down the pike in Q1 or Q2.

this will be fun to watch.

 

[incandescent]

oh goodie! maybe they will put an even bigger HSF on it!

 

[ZoinKs!]

My understanding of the $400 million dollar number was that it's for the entire nv3x line. Everything, including the nv30, mx and mobile versions, and the nv35. However, I have also suspected that early nv30 designs were either scrapped or heavily reworked. I think it was Dave Kirk who said that it took a lot of time to get the nv30 to work right.

 

[Pete]

so in all likelyhood, Nvidia should not have NEARLY the same problems with NV35, that it did with NV30. provided they don't have trouble with the NV35 itself. The .13 micron process problems are all ironed out now, or will be shortly. I don't see why Nvidia would not have a refreshed core with some enhancments + clock increase, by June-July. they almost have to with R350 coming down the pike in Q1 or Q2.

this will be fun to watch.

Someone on Ars pointed out that nV was having trouble with TSMC's low-k dialectric .13u process, and they switched to the hotter-running .13u process, which resulted in the delay and the extra power and cooling requirements evident on the 5800U. So nV may still have some kinks to work out on the .13u process that can save them more power.

 

[Gollum]

As various sources said though, NV30 specs have been set in stone and basically not changed at all since the initial paper-specs were sent out in the February timeframe. I'm not saying that nothing was changed, but all these rumors of adding more pipelines (and before that adding a 256bit bus) seem highly unlikely to me as a reason for the delay...

 

[Joe DeFuria]

Regardless, the NV30 is a huge chip, and don't forget that nVidia has been getting samples back from TSMC on .13 micron since Februrary.

I never heard that. Source?

I have heard that nVidia have been working with / testing 0.13 and have been getting some kind of "test" 0.13 silicon back since feb. But I haven't heard anything about having NV30 samples at anything close to that time frame.