Will the 7800 "Ultra" be 90nm ?

[Zenoth]

Topic title. Question asked.

I'd like to know. I haven't been around looking for news related to GPU's lately. But I've read threads about it today, here at Anand's forums.

Very interesting GPU. But I was wondering ... if nVidia decided just to put the frequencies on steroids and that's it, or did they actually changed something "here and there" other than frequencies and the actual amount of Memory (now 512 MB) ?

I'll get my X1800 XT next week.

But if that nVidia card is interesting enough to me, I might sell my current GTX, and buy that "Ultra" version and just pay the difference.

 

[xtknight]

I really doubt it.

 

[Elfear]

Originally posted by: xtknight
I really doubt it.

:thumbsup:

With as much growing pains as ATI had, I imagine Nvidia will be in the same boat when they switch over.

 

[Zenoth]

I see.

Hmmm. So they pushed it that far on 110nm. Most impressive.

I'm still curious, and thanks God, patient as well. I think I'll just wait for R580, and then decide which one would be best suited for my tastes.

I bought my Leadtek only three weeks ago. So the announcement of that "Ultra" version isn't something I like at the moment. I expected it, but not already.

 

[BFG10K]

Unlikely.

With as much growing pains as ATI had

Actually ATi's issue turned out to be quite trivial, it just took them a long time to find it.

 

[1Dark1Sharigan1]

Next gen nvidia probably will be, but not the refresher part . . . nvidia already knows its getting great yields out of its 110nm process so suddenly switchin to 90nm for a 7800 refresher part is just short of insanity . . .

 

[BenSkywalker]

With as much growing pains as ATI had, I imagine Nvidia will be in the same boat when they switch over.

ATi doesn't fab their own chiips, TSMC does it for them. TSMC also fabs nVidia's chips. nVidia seemed to learn from their mistakes of being the leader on a process change, ATi took the hit this time around. Now that TSMC has gotten a decent amount of time under their belt with 90nm and worked out a lot of issues, nV's transition is certain to go far smoother. Not saying they will or won't have an Ultra @90nm, just that 90nm @TSMC is obviously a far more mature process now then it was when ATi started working on their part.

 

[cronic]

Originally posted by: BenSkywalker

With as much growing pains as ATI had, I imagine Nvidia will be in the same boat when they switch over.

ATi doesn't fab their own chiips, TSMC does it for them. TSMC also fabs nVidia's chips. nVidia seemed to learn from their mistakes of being the leader on a process change, ATi took the hit this time around. Now that TSMC has gotten a decent amount of time under their belt with 90nm and worked out a lot of issues, nV's transition is certain to go far smoother. Not saying they will or won't have an Ultra @90nm, just that 90nm @TSMC is obviously a far more mature process now then it was when ATi started working on their part.

i think the 90nm bomb will be dropped on the r580

 

[KeepItRed]

Originally posted by: cronic
i think the 90nm bomb will be dropped on the r580

True, I think the R520 was just an experiment for 90nm technology. I can picture a R580 with all those pipes and whipped cream on top.

 

[nRollo]

Originally posted by: BFG10K
Unlikely.

With as much growing pains as ATI had

Actually ATi's issue turned out to be quite trivial, it just took them a long time to find it.

I don't believe their "one peripheral circuit" story, I think they're just telling us that because they don't want to admit three re-spins to get it to current clocks.

That said, I doubt they consider three re-spins, months of delays, and the millions of dollars lost "trivial".

 

[Zenoth]

because they don't want to admit...

In politics, truth isn't lucrative.

And it's usually the same thing in marketing, or even engineering.

Saying they noticed that they had to change their priorities, or to find an "issue", is already a good move from them. But actually saying they indeed had technical troubles due to instability, or else, then it would have been considered a "weakness" in consumers' eyes.

After all, who wants to buy a product from a company that "admits" there has been "x" issues on the said product before the release ?

Not many.

 

[Gstanfor]

Originally posted by: Rollo

Originally posted by: BFG10K
Unlikely.

With as much growing pains as ATI had

Actually ATi's issue turned out to be quite trivial, it just took them a long time to find it.

I don't believe their "one peripheral circuit" story, I think they're just telling us that because they don't want to admit three re-spins to get it to current clocks.

That said, I doubt they consider three re-spins, months of delays, and the millions of dollars lost "trivial".

The fault was in a third party library - ie: Fast14, the technology that the fanbois loved to crow about, saying nVidia had nothing to match it. Little did they realise (till I pointed it out to them) that nvidia had been using Arithmatica's Cellmath libraries since NV40 first launched :laugh: :laugh: :laugh:

The R5xx series has been good for a laugh if nothing else!

 

[Cookie Monster]

The G71/G72/G73s will be all 90nm. The midrange lowrange cards are all 90nm.
According to Nvidia, the 90nm die shrink will allow them to produce 2 GPU SLi on single cards, using similiar size for the PCB.

Anyways, according to the INQ, some GTX 512mb cards can clock up to 800mhz. Amazing, but i think this G70 core is tweaked from the original.


Edit-The ultra probably is 110nm. Ill be surprised if it is 90nm. And 8 quad. 32 pipes anyone?

 

[Pete]

Sources at B3D say the upcoming 512MB GTX/whatever is still 110nm, but a respin of the original G70 (metal and layer changes, IIRC). I'm surprised to see a 110nm non-low-k part clock so close to ATI's 90nm low-k part, but obviously we're not comparing apples to apples, and G70 is a lot more mature than R520, seeing as it's a closer extension to NV40 than R520 is to previous PC parts.

Greg, was ATI's publically admitted problem from Fast14 itself, or someone else?

 

[nRollo]

Originally posted by: Pete
Sources at B3D say the upcoming 512MB GTX/whatever is still 110nm, but a respin of the original G70 (metal and layer changes, IIRC). I'm surprised to see a 110nm non-low-k part clock so close to ATI's 90nm low-k part, but obviously we're not comparing apples to apples, and G70 is a lot more mature than R520, seeing as it's a closer extension to NV40 than R520 is to previous PC parts.

Greg, was ATI's publically admitted problem from Fast14 itself, or someone else?

Don't companies mainly due die shrinks with their mid-range parts these days to avoid problems like nV encountered with the nV30?

I know ATI broke the pattern with the XT, but imagine they had to to get the clock speed to compete.

 

[Pete]

The immediate NV30 aftermath may have been a cautionary detour, or maybe even a forced one, if TSMC was having that many problems. ATI said they needed 90nm for what they wanted to do with R520, so it might not have been just clock speed.

But, aside from NV30, NV rode the smaller-process wave to the top of their field. They did it for good reasons (more chips per wafer and higher clocks per chip), so you'd think they'd continue ASAP. ATI apparently had technical in addition to the usual long-term financial reasons to wait for 90nm.

I guess the other reason we've seen smaller parts go to smaller processes first is b/c the former are in greater demand and the latter have greater capacity. IHVs may want to keep their top end stable for as long as possible simply b/c they may not sell as many, and so retapes or die shrinks may not be as attractive.

 

[Gstanfor]

Originally posted by: Pete
Sources at B3D say the upcoming 512MB GTX/whatever is still 110nm, but a respin of the original G70 (metal and layer changes, IIRC). I'm surprised to see a 110nm non-low-k part clock so close to ATI's 90nm low-k part, but obviously we're not comparing apples to apples, and G70 is a lot more mature than R520, seeing as it's a closer extension to NV40 than R520 is to previous PC parts.

Greg, was ATI's publically admitted problem from Fast14 itself, or someone else?

No names have been mentioned by anyone, however it isn't difficult to add 2+2 & get 4 in this case. There have been several reports around the net of designers finding Fast14 very tricky to use in real designs effectively.

 

[Pabster]

Originally posted by: BFG10K
Actually ATi's issue turned out to be quite trivial, it just took them a long time to find it.

Heh, talk about looking for a silver lining.

A "trivial" problem doesn't delay release to market for over 6 months.

 

[xtknight]

Originally posted by: Pabster

Originally posted by: BFG10K
Actually ATi's issue turned out to be quite trivial, it just took them a long time to find it.

Heh, talk about looking for a silver lining.

A "trivial" problem doesn't delay release to market for over 6 months.

AFAIK it didn't even have to do with the process. It took them very long to find a bad gate.

 

[OneOfTheseDays]

i think it's funny how people can comment about what the cause for ATI's delay was when they have absolutely no engineering background whatsoever. if the problem was indeed one faulty gate, that's the equivalent of looking for a needle in a haystack.

 

[Dethfrumbelo]

I remember having to design some fairly simply logic in semiconductor lab, involving a couple hundred ANDs/ORs or so, and then do the litho/etching/doping. I think I was trying to make a multiplexer or something. It never really worked correctly, and most students couldn't get theirs to work either, whether because of bad gates or design. Multiply that complexity by a billion or more and that's what they have to deal with. Not easy.

 

[xtknight]

Originally posted by: Sudheer Anne
i think it's funny how people can comment about what the cause for ATI's delay was when they have absolutely no engineering background whatsoever. if the problem was indeed one faulty gate, that's the equivalent of looking for a needle in a haystack.

It said it right on a reputable hardware review site, and I'm sure ATI were the ones that told the reviewers that. It's not like they're spouting off nonsense. Besides, even if I did have an engineering background, how would I know what their problem is? That doesn't make you psychic. It seems like a perfect explanation for the long wait to me.

http://www.guru3d.com/article/Videocards/262/

I originally read it some where else also but I can't remember where.

 

[jiffylube1024]

Originally posted by: Pabster

Originally posted by: BFG10K
Actually ATi's issue turned out to be quite trivial, it just took them a long time to find it.

Heh, talk about looking for a silver lining.

A "trivial" problem doesn't delay release to market for over 6 months.

You guys aren't playing fair with words. The problem itself turned out to be trivial. It's effect on ATI (6 months of delays, doublespeak, hearsay, losing face) was not trivial.

 

[Gstanfor]

Beyond3d Exerpt

According to public reports ATI noticed that as late as July, issues occurred that prevented the R520 core being clocked close to its target speeds, which is consistent with leakage issues. Curiously, the issue did not occur across all their 90nm products - ATI had already delivered Xenos to Microsoft using the same 90nm process R520 does, and other derivatives of the R520 line suffered the same issue (RV530) but others did not (RV515) - the fact R520 and RV530 share the same memory bus, while RV515 and Xenos have different memory busses is not likely to be coincidental in this case. ATI were open about talking about the issue they faced bringing up R520, sometimes describing the issue in such detail that only Electronic Engineers are likely to understand, however their primary issue when trying to track it down was that it wasn't a consistent failure - it was almost random in its appearance, causing boards to fail in different cases at different times, the only consistent element being that it occurs at high clockspeeds. Although, publicly, ATI representatives wouldn't lay blame on exactly were the issue existed, quietly some will point out that when the issue was eventually traced it had occurred not in any of ATI's logic cells, but instead in a piece of "off-the-shelf" third party IP whose 90nm library was not correct. Once the issue was actually traced, after nearly 6 months of attacking numerous points where they felt the problems could have occurred, it took them less than an hour to resolve in the design, requiring only a contact and metal change, and once back from the fab with the fix in place stable, yield-able clockspeeds jumped in the order of 160MHz.

Fast14 has to be customized to every process you apply it to, on a fab by fab and process by process basis. It can be difficult to design good circuits easily and Fast14 is power hungry to boot.