"Big" Maxwell GM200 Has Been Spotted!

[SinOfLiberty]

As of today,

via zauba.com

 

[3DVagabond]

22,245 inr = $370usd

 

[HurleyBird]

The big question is... 20/16nm?

 

[3DVagabond]

28nm

 

[SinOfLiberty]

20nm

 

[3DVagabond]

Sorry guys. I completely missed it's GM200, not 204. I wish AMD posted their new chips to India so we could see the shipping manifest.

 

[Cloudfire777]

22,245 inr = $370usd

Just for the die itself.

 

[3DVagabond]

Just for the die itself.

Yes. That's the insured amt. I assume. I'm not sure exactly what the actual value is as a component for purchase. If it ends up they actually sell the raw chips for that much it's going to be an expensive card.

 

[tviceman]

I'm calling FAKE on this one.

 

[Mand]

I'm calling FAKE on this one.

Who would fake a zauba shipping list and why?

 

[3DVagabond]

Who would fake a zauba shipping list and why?

Someone who wanted to leak it onto the web, generate page hits, get off trolling everyone. Lot's of people.

 

[tviceman]

Who would fake a zauba shipping list and why?

The same person / people who would fake specs for upcoming cards or creating entirely non-existent cards with a company's marketing colors and fonts in a power point slide.

GM100/110 cannot feasibly exist on 28nm whilst simultaneously offering the same performance jump over it's predecessor that GM107 offered over GK107. So if it does exist it will be a disappointing new chip for it's size, likely only performing slightly faster than GM204 in graphics or it won't be released as a graphics chip at all.

Just my 2 cents.

 

[FatherMurphy]

Is 600 mm^2 even possible at TSMC on 28nm? See the 3dCenter chart here, discussing GM200: http://www.3dcenter.org/news/nvidias-gm200-chip-erstmals-gesichtet.

I want to say that GK110 maxed out the die size at TSMC on 28nm (though perhaps they have increased the max size to accommodate the drawn out life cycle of 28nm).

 

[96Firebird]

If it ends up they actually sell the raw chips for that much it's going to be an expensive card.

Production chips won't be near that much. These engineering samples are ordered as needed, one wafer at a time. I'm actually surprised they are insured so cheap...

 

[SinOfLiberty]

This is not the 1st time GM200 is spotted. So someone faking the zauba list is nonsence, as this gpu does exist. Es samples fly around a few months before the release.

This is of course an ES but from the taken info, GM200 is being mass produced at this point.

 

[Cloudfire777]

Is 600 mm^2 even possible at TSMC on 28nm? See the 3dCenter chart here, discussing GM200: http://www.3dcenter.org/news/nvidias-gm200-chip-erstmals-gesichtet.

I want to say that GK110 maxed out the die size at TSMC on 28nm (though perhaps they have increased the max size to accommodate the drawn out life cycle of 28nm).

Of course it is possible.
We soon have a GM204 on approx 430mm2 beating a the big 561mm2 GK110 by 20%-50% (not known what yet).
GM200 with same die size as GK110 (561mm2) will have a lot more cores than GM204 and the performance likewise will be much better than GK110.

 

[FatherMurphy]

I am not saying that GM200 or GM204 can't top GK110 in performance. If NVIDIA brought out either that didn't top GK110 in performance by some margin, I'd be baffled.

I'm saying that TSMC's 28nm process can only make die sizes so large (i.e. the reticle size of the lithography machines). See this quote from PCPER discussing GK110

The GK110 is a very large chip at 7.1 billion transistors. It likely is approaching the maximum reticle limit for lithography, and it would make little sense to try to make a larger chip. So it seems that GK110 will be the flagship single GPU for quite some time.

http://www.pcper.com/reviews/Graphi...10-GPU-Boost-20-Overclocking-and-GPGPU?page=1

So, if GM200 is on 28nm, I would not think that it would be any bigger than GK110's 561mm^2. 3dCenter is saying the die size of GM200 would be approxiately 600 mm^2.

I think any performance increases over GK110 need to happen in the same amount or less die space if 2nd gen Maxwell remains on 28nm.

 

[Mand]

I am not saying that GM200 or GM204 can't top GK110 in performance. If NVIDIA brought out either that didn't top GK110 in performance by some margin, I'd be baffled.

I'm saying that TSMC's 28nm process can only make die sizes so large (i.e. the reticle size of the lithography machines). See this quote from PCPER discussing GK110

The GK110 is a very large chip at 7.1 billion transistors. It likely is approaching the maximum reticle limit for lithography, and it would make little sense to try to make a larger chip. So it seems that GK110 will be the flagship single GPU for quite some time.

http://www.pcper.com/reviews/Graphi...10-GPU-Boost-20-Overclocking-and-GPGPU?page=1

So, if GM200 is on 28nm, I would not think that it would be any bigger than GK110's 561mm^2. 3dCenter is saying the die size of GM200 would be approxiately 600 mm^2.

I think any performance increases over GK110 need to happen in the same amount or less die space if 2nd gen Maxwell remains on 28nm.

I still don't understand why people think there is some cataclysm that has swallowed up 20nm, never to be seen again.

 

[Cloudfire777]

Yes, like I said above, they can stay at around 550mm2 and deliver a really fast Maxwell product. They need less Maxwell cores (and space requirement) to beat Kepler since 1st gen Maxwell deliver 35% better performance per core than Kepler. That may even increase with 2nd generation (GM2xx).

I don`t think they have ever built a bigger die than 560mm2 anyway, and they dont need to.

 

[r3pshow]

On tweaktown they said, there's also a GeForce GTX TITAN II coming ! :|

Link

 

[n0x1ous]

Yes, like I said above, they can stay at around 550mm2 and deliver a really fast Maxwell product. They need less Maxwell cores (and space requirement) to beat Kepler since 1st gen Maxwell deliver 35% better performance per core than Kepler. That may even increase with 2nd generation (GM2xx).

I don`t think they have ever built a bigger die than 560mm2 anyway, and they dont need to.

Original 65nm GT200 was 576mm2

 

[FatherMurphy]

I still don't understand why people think there is some cataclysm that has swallowed up 20nm, never to be seen again.

No one said that. It's just that the reports we are all speculating on right now (via 3dCenter) state that this supposed GM200 is 28nm.

 

[Cloudfire777]

I still don't understand why people think there is some cataclysm that has swallowed up 20nm, never to be seen again.

20nm may be a SOC process only for Altera, Qualcomm and Apple.

16nm FinFET is still 20nm just better in every way. Thats why TSMC seems to be quick at getting it up and running. Samsung & GlobalFoundries, Intel, they are all breathing in their necks, and if TSMC want to keep their clients, they better step up (with FinFET)

Original 65nm GT200 was 576mm2

550mm2, 560mm2, 570mm2, pretty much the same. I`m not sure where the limit goes though.

 

[SinOfLiberty]

I still don't understand why people think there is some cataclysm that has swallowed up 20nm, never to be seen again.

Jimmy: -Dude, man no way 20nm is making it into gpus.

Mike"-Why?

Jimmy-Cause some (enter name of any news site) has reported so..

Mike: They saying it based off of what?

Jimmy:- Secret sauce!

Mike:-:hmm:--> :awe:


20nm has never gone anywhere. It just more expensive to make due to Soc orientated competition.

 

[Mand]

No one said that. It's just that the reports we are all speculating on right now (via 3dCenter) state that this supposed GM200 is 28nm.

But the only reason they're saying that is because people have been saying for months that Maxwell won't be 20nm, because...uh....YEAH!

There was one report of an issue with 20nm that caused a minor delay, one report that TSMC was very happy to be making 20nm SoCs, and then people assumed that meant no 20nm Maxwell. Then another site talked about something else and said "and since we think there's no 20nm Maxwell." Then another, then another, then another.

Now, it has echo-chambered itself into Something We All Just Know, even though there is no source for it whatsoever. So every bit of news is getting filtered through assumptions that there won't be 20nm Maxwell. It's just become an ingrained part of the discussion.

I wouldn't be surprised in the slightest to see a 20nm GM204 this fall.

Maxwell has always been planned around 20nm. Do you really expect Nvidia to abandon that plan just because of one delay that was fixed quickly?