NV30 not taped out yet....

I guess I'm not sure exactly what it means.

Could be time from initial tapeout to the final tapeout thats good enough for production. The times seem to fit (approximately) assuming each metal rev takes about 3-6 weeks on average to get back from the fab, evaluate, and come up with the fixes.

 

That's right. But also wrong. They did launch the GF3 with bug intact and later corrected the problem.

110 days from taping out say next week would give the product launch at about November time if NVIDIA are able to execute from the initial tapeout to end production without any further delays and without requiring any new revisions. That is a very optimistic scenerio however - but it is not something I would bet against.

This is why I asked how long it is between revisions. Otherwise, as some have mentioned, the data is slightly useless for guestimating where NVIDIA is right now with regards to a NV30 launch.

 

thanks for the info

 

Indeed. The A12 stepping is the second incarnation of R300 (I think R200 ramped using the A13 mask set). The respin took about 1 month and two weeks plus the eng time it took to fix the memory controller bug.

MuFu.

 

ATi's differs from NVIDIA's - I asked when R200 first came about. That went to A13 and that was 3 spins - so they basically use 1 where NVIDIA use 0.

 

ATI starts at A11. The R200 went through 3 revisions with the A13 revision released to the public. The R300 will be released at A12 and I imagine ATI has been able to keep their production window close to 100 days in terms of initial tapeout to volume ramp.

 

On the shareholder broadcast the CEO was very clear in his statements here, not once did he say Would be ready for the holiday season but instead said Should be ready...
To me that is a clear indication there is still no working Silicon.

 

About the GF3, did you take in acount the numbers of days it took from is real launch till we could buy products in volume? As there was a long gap.

ALso I dont expect them to get any better at going from tape out to production. They may get worse as the designs get more and more complexed, the more things that can go wrong...

 

I concur & add that never have I seen so much flip-flopping info passed from partner IHVs & OEMs... The 128bit databus & 16bit fp pipeline numbers were written in stone as little as 6 weeks ago...

The cash cows for the chain are the GF4MX series esp 420 & 440. The 460 is basically a "free" markup... GF2MXs continue to sell well esp in 64MB SKUs.

 

Le Stoffer, I am shocked that you actually still think he was talking about final silicon. If he was, he would have been more confident for a launch this year. If they are that close to FINAL SILICON, then they would have product on shelves in less than 2 months.

This has been beaten like a dead horse but one more time, he was talking about FIRST SILICON. And, he is hoping it is perfect in order to ship product this year.

The reason why people thought that they had taped out back in June is that Nvidia was planning on 'revolutionizing the GPU market' in August. Thus, they would have silicon to announce NV30. Reality is, they must have been planning on a late July, early August tape out, thus then starting the PR machine for NV30.

 

Has anyone thought that maybe the Nv-30 did have final silicon but they just made some last minute changes in these last couple of months.{e.g 256bit bus}.Since he did say over here that a 256 bit bus was overkill & I remember reading somewhere that TSMC had already addressed the issues with the new process & this seems to be the only logical alternative of adding more to an already crowded die,imo.

 

If they already had final silicon it would be a tremendous waste of money & time to scrap it in order to make major revisions to it. From what I understand, its not exactly just changing a few numbers in a computer . More likely they would've released the card then refreshed it with improvements later.

 

San Greal, you are correct. They would NOT scrap a finished product.

 

Actually, there's only time enough for about 1 more metal rev (taping out in the next week or so) before you cannot physically get product out in time for oems to build and ship by mid December.

Supposing they get the next rev back in a month, it verifies in 2 weeks. Any fab takes 1.5-3 months to go from wafer start to packaged parts, right now, fabs are full so its closer to 3 months. Add in a month of slop and we're at the knife edge.

So, I'd hazard a guess the CEO was talking about the final revision tapeout, if he was really serious about making the Christmas market.

Strange terminology, though.

 

Unless he wanted people to sell his stock he would have put things in the best light possible. Believe me, if there had been a preliminary tape out, he'd have said so. The fact that he was forced to admit there had been no tape out basically means its the worst case scenario (since no other words could be used to describe the situation).

 

I seriously doubt that any delays would be due to revisions in the products' design.

Instead, we may have revisions in the NV30's design as a result of delays that came from lack of a mature .13 micron fab process.

 

Revisions also occur because of bugs that can't be worked around.

 

Of course, and I'm sure there have been those, too.

But, I don't think that any reasonable number of bugs (which always will occur) would delay the NV30 past a September release.

It should be apparent that the #1 reason that the NV30 isn't going to be available in September is because the .13 micron process is still immature.

 

I don't understand this. If the chip hasn't taped out yet, how can you expect a September release? Even if everything goes flawlessly, it still takes over a month to get chips back. Then you have to make boards out of them. Test the boards you've built. Get the driver WHQL certified. Etc. Etc. There's much more involved than you are considering.

How can you blame everything on the process if the chip hasn't even taped out yet? You don't always know the process is an issue until you actually try to build some chips.

 

I might actually disagree with you there Chalnoth.

What if they were ready to almost tapeout(remember that they did initially say that the NV30 would be released in September/October can't remember which now) but heard through the grape vine that ATI would release a 256bit bus where thye would only have a 128bit bus .. it would make sense for me if they made calculations and came up with the lower bandwidth calculations(and performance) than the R300 and actually stopped the actual tapeout from happening and begun working on a 256bit Bus. This could be the one reason why they are taking longer. I do think Nvidia might have been ready .. but had to reconsider either bringing out a slower card or a quicker card 1-2 months later.

This would bring into question.. how much quicker is the NV30 compared to the R300. Well lemme say this.. when the GF3 was initially released it was actually slower than the GF2 Ultra. Only after games, software and drivers came out did the GF3 actually overtake the GF2 Ultra in performance. Also the fact that they came out with the Nvidia nfiniteFX™ Engine and Lightspeed Memory Architecture which did improve the performance of the card over the Ultra fast GF2.

So I can see that although the card might not be initially faster that the R300, it woould make sense that when games/software came out using DX9 or CineFX or Nvidia CG instructions .. that is the time we will see the card move into a new era and thus beating the R300 hands down.

US