NVIDIA Kepler speculation thread

What makes you say that?

 

Why do you need a big event to launch a new GPU? Did AMD use a special event to launch 7970?

These things take years to develop. Inevitably somebody is going to be first. It doesn't matter that much, even if there's a quarter difference. That's especially true these days with the lack of half nodes, which stretches out the time between major jumps in performance.

Both Nvidia and AMD should be able to ride the 28nm for a good 2 years.

 

Please refresh my memory on that one. I don't recall nVidia ever releasing a better product on the same day that ATi/AMD released one of theirs. Yes, there have been times not long after a release that nVidia released a far superior product... but I don't believe nV has any desire to see AMD fail to the point of losing their competitor and protection from anti-trust legislation.

AMD has never been the most sound company when it comes to finances and stability. And truth be told, since the companies founding 42 years ago.... they are still at a Cumulative Net Loss over all those years combined.


Edit: I just noticed I'm listed as a "junior member" even though I've been here for TEN YEARS!!!! wtf?
http://www.forbes.com/sites/michaelkanellos/2011/11/07/where-amd-failed/

 

If I remember correctly, the last one was pushing GTX460OC models for review at the time of Barts launch. There are more such instances.

No demos at this show is neither conclusive or inconclusive of anything IMO.

 

Isn't May too late?

 

Lol, they're certainly not early then.

 

If an actual product got released it wouldn't be a spoiler would it? Nvidia usually issue white papers about their upcoming products, or "accidentally" leak internal memos slamming the opposition, or instructions on how new competitor cards should be reviewed against their current ones. Lots of stuff through the archives here if you care to search.

NV is the king of PR, and rarely let an opportunity pass to dress down the opposition and boost themselves up, but they have been unusually quiet against the first 28nm GPU claiming the world's fastest crown.

 

Probably Cuda Core of Kepler will be set to (fp*2 int*1).

http://www.obr-hardware.com/2012/01/exclusive-28nm-geforce-specs.html

 

Well if my memory serves me right, the above events normally took place when they didnt have anything to respond with against the competition.

I mean, theres been several times nVIDIA launched succesfully without any buzz noise/leaks at all took everyone by surprise. G70, G80, GT200 are some of the launches that come to mind.

 

old stuff but this time for kepler lol..

http://forum.beyond3d.com/showpost.php?p=1470740&postcount=6613

anyway another waste of bytes


http://translate.google.com/translate?langpair=auto|en&u=http%3A%2F%2Fascii.jp%2Felem%2F000%2F000%2F658%2F658986%2Findex-3.html

 

Probably

CudaCore[ fp*2,int*1 ]
SM[ CudaCore*48,TMU*8 ]
GPC[ SM*3 ]

GK104[ GPC*4,ROP*48 ]
GK106[ GPC*2,ROP*32 ]
GK107[ GPC*1,ROP*16 ]

 

I find it hard to believe that you're the only one in the world who didn't read the leaked AMD review guide for the 7970.

 

A 3 month or even 6 month gap between nVidia and AMD by itself doesn't matter - if the end product is exciting. Unfortunately, that kind of delay is usually accompanied by a disappointing product (R600 and to a lesser extent Fermi). That tells me that the extra time is usually spent trying to fix things and not necessarily because they were aiming higher than the competition and planned to take longer getting there.

I welcome evidence to the contrary but Kepler was probably targeted for 2011 and is late based on nVidia's internal schedule. The real kicker is why it's late.

1. Manufacturing problems (Fermi)
2. Architectural problems (R600)
3. Poor 28nm yield at TSMC?
4. Tactical decision?

 

Precisely. You can't win the game if you don't play. By saying/showing absolutely nothing while their competitor has announced and released a product to generally positive reviews they are simply giving up money. By remaining utterly silent they are giving the impression that something is seriously wrong or really late and people who are ready to upgrade have a choice between (A) a good product that is currently available (sort of) or (B)"?".

 

IMHO 3 and 4. SI was targetted for 2011 too, at least that's what AMD publicly stated more than once.

And I don't necessarily agree to your examples in 1 & 2 either. Fermi had 1 or more hw bugs, but it wouldn't had been manufacturable in late 2009 due to crappy yields and in extension too high costs anyway and it's not like R600 had a joyride with 80nm@TSMC which NV for some weird reason avoided completely.

Since 28nm doesn't sound so far like any sort of worthwhile improvement in terms of wafer yields compared to 40G I have severe doubts that NV could have manufactured their top dog as early as 2011 either. It's my understanding that their original plan for Kepler was for quite some time to go with the performance part first and later on to go for the top dog.

The question now is when NV will launch its GK104. If it's quite a few months after Tahiti then it obviously can't be a yield issue, but something else. Until the real release date is known I doubt we'll find out what is really going on. The internet is at the moment going through its typical silly season with idiotic fake diagrams with conflicting data appearing left and right. One message that might be of some value is "earlier than expected". So when was GK104 really expected after all, since no next generation 28nm GPU made it on shelves in 2011?

 

What money? It's early January after an expensive and broadly speaking unaffordable Christmas (to most normal people in the Western world with a mortgage and kids at least). NV's competitor missed Christmas too, and now nobody has any money left to spend on expensive graphics cards.

When is the next properly important back-to-whatever OEM cycle?

 

In some parts never but that's besides the topic here

 

Yes, that's definitely the impression it gives. You could fabricate scenarios where nVidia's silence isn't a cause for concern though if you were feeling optimistic.

A combination of poor 28nm yields, lower target pricing and competitive performance would explain things as well. They have no reason to spoil AMD's party now if they expect to make a move soon. A trickle of 7970's isn't a threat. A far sweeter victory would be to get revenge for what Cypress did to GT200.

Having said that, I'm a firm believer in Occam's razor!