NVIDIA Kepler Inbound for March-April

[Casecutter]

If there is anyone who believes Nvidia will release their Über offering and be able to set an MSRP lower than AMD 7970, I want what you're smoking!

Kepler will be an improvement over the 7970 in performance, but honestly not on power or production costs unless they're really able to pull a rabbit out of the hat. So, it still does mean complex power sections and coolers to engineer and package onto a PCB.

Per chip I can't see TSMC giving any price break to Nvidia. The only way this works is that if Nvidia purchases the entire wafer and their architecture sort much better and they get better yields. I don't know how either group contacts on purchasing or that either side has an architectural advantage in production at 28Nm. They’re both probably struggling with TSMC, the only upside is Nvidia will be reaping production benefits for being behind, while AMD has 5mo's ahead in sales. It probably just evens out a on the bottom line by year’s end.

 

[Benetanegia]

If there is anyone who believes Nvidia will release their Über offering and be able to set an MSRP lower than AMD 7970, I want what you're smoking!

I don't know why would you say that. GTX580 is already as fast per clock as HD7970. The only thing that Nvidia needs is a die-shrink. But Kepler is more than that and arguably much better/efficient than Fermi. Unlike AMD which had to add almost everything GPGPU related, Kepler only needs some incremental updates on that front, because Fermi already had 95% of them. That will hardly mean a decrease in gaming efficiency. There's a penalty for going to GPGPU, but unlike AMD Nvidia has been paying it since G80 days.

Unless they screw up royally, maybe by trying to introduce a lot more GPGPU features than needed (unlikely) or changing the architecture entirely 2 times in a row (not gonna happen ever), Nvidia will most probably have the most efficient chip this time around.

They will definitely be able to set a lower MSRP on their competing products, whether they want to do it or not, that's another story. If rumors/marketing is true and GK104 (performance part) can compete/beats HD7970 then they will most definitely set a lower price for it. The $500-600 spot wil be reserved for the 2-3 GK100 SKUs. There's simply not a lot of room for a single GPU card above the $550 that AMD already asks for. IMO Nvidia will price GK104 at around $400, and with a smile in their face due to the huge profits. I'm 95% sure they were expecting a better performing and far cheaper HD79xx, in the end GK104 was supposed to be GF104's replacement in the market, so $200-300 was most probably in Nvidia's mind for the chip.

 

[OneCool]

According to the report, NVIDIA wants to avoid doing a paper-launch like AMD

Stupidest shit I ever heard.Like they have never done that before

 

[Casecutter]

They will definitely be able to set a lower MSRP on their competing products,

Well, we wait and see, but this is not going to be like when the GTX260 first showed 06-16-2008; then like 10 day later the 4870 hit market and the price of those GTX260's went in to free fall.

With this news and what you said Nvidia really needs the GK104 to get here quick if they want in on a piece of the action, April is an eternity away. The last time folk held out (Fermi) it play out opposite to the early prophets, wait another month or two and we get something tangible. I'll buy into what your presenting, but as of yet we've got nothing (tangible). Even at CES when Jen Hsun Huang was asked, where’s Kepler? The answer was, “we have to be "patient" about it”.

And again price is based on chip production and whatever TSMC is getting as their new (increased) price on 28Nm. AMD is in the market floating a price based on that new structure, we need to see how it floats when Nvidia cracks to bottle over their bow.

Would I like Nvidia to get a 7950 buster at $350… that would be a good play, and competition to the rivalry. I just don't know if they'll be that agressive. While by the end of March AMD might know the writing's on the wall, lowering the price after 2-3mo's in the market, and we get a real horse race in the $300-350 market.

 

[ensabrenoir]

Patient?????

Hope patient doesnt turn into we got no competive answer to Amds product.

 

[xenocide]

Stupidest shit I ever heard.Like they have never done that before

I personally hate paper launches. Knowing how great a product is doesn't matter when you won't be able to buy it for several weeks, and even when you can it is in low supply.

I think Kepler will be very competative with AMD offerings.

 

[Benetanegia]

The thing about being patient can also mean "be patient, it is close and we will reveal it soon (don't spoil it dumbass)". It could mean anything really.

I hope you have your popcorn. I found these:

http://vr-zone.com/articles/rumour-nvidia-geforce-gtx-680-to-release-in-february/14552.html
http://www.obr-hardware.com/2012/01/kepler-is-on-track.html

Both rumors saying practically the same. Launch in February. Still rumors, maybe true, maybe not...

I do know that if I was Nvidia and had a winner, I'd try to hide it until launch for a greater impact. Like they did with 8800 or like AMD did with HD48xx launch. Something like HD48xx/GTX2xx is not very likely to happen ever again, but I did say in another thread that IMO AMD pushing and pulling HD7970 and 7950 release dates around the way they did was a defense against that precisely. While only rumors, pretty much the same can be said about Kepler.

IMO this is what has happened lately:

1- AMD hears of Kepler (GK104) being at TSMC already, for production.
2- AMD pulls release date of HD7970 to December, to maximize sales at $550. HD7950 on Janary 9.
3- AMD gets more concrete info about Kepler. Specifically that it might be very competitive with Tahiti.
4- AMD delays HD7950 as a defense, no concrete release date, but February is the most likely option. Nvidia is going to talk about Kepler at CES according to non-confirmed schedule, so after knowing what they will be facing, a date and a price will be set for Tahiti Pro. Nvidi won't play a HD48xx on them.
5- Nvidia cancels any metion of Kepler at CES. When asked about why Huang only says to be patient.
6- Nvidia says they won't release Kepler until they have plenty of stock. The media estimates release date to be April.
7- Without any info on what exactly Kepler is, there's no benefit waiting anymore, so AMD pulls HD7950 release as much as they can, which happens to be the end of January.
8- (?) With the spy game pretty much over now, Nvidia pulls the release as much as they can, which is late February/March.

 

[THE_EGG]

Yeah it's certainly doable, but I wear glasses, don't have the perfect vision, and I'm behind a PC a lot (software tester), so I do enjoy when I don't have to go closer to my monitor to read my desktop text

+1 I hear that!

 

[RejZoR]

FXAA was made by Nvidia. Instead of being hardware locked (i.e. Nvidia only cards like PhysX) however, they left it to the game developers to add it to the game.

Well what good is it that if developers don't even bother adding any FSAA to the PC games? AMD's MLAA can be enabled in any game. If it has FSAA or not, you can use MLAA if you want to. You don't have to rely on anyone to enable it for you. That's why i'd like to have it as user controled setting rather than leaving it to the devs who clearly don't give a damn.

 

[ViperXTR]

its already inside nVidia drivers, FXAA that is, it remains hidden until now but it can be enabled and tweaked, dunno why its still hidden (it was there for several weeks/months now since FXAA was introduced), maybe when the kepler is out, nVidia will finally put it in the main control panel.

Oh and Framelimit is also available, an alternative to vsync without tearing, supposedly.

 

[entropy13]

Well what good is it that if developers don't even bother adding any FSAA to the PC games? AMD's MLAA can be enabled in any game. If it has FSAA or not, you can use MLAA if you want to. You don't have to rely on anyone to enable it for you. That's why i'd like to have it as user controled setting rather than leaving it to the devs who clearly don't give a damn.

Except MLAA doesn't work with the competitor's cards, i.e. Nvidia.

And it's FXAA, not FSAA.


I'm basing my info from here.

 

[Casecutter]

The thing about being patient can also mean "be patient, it is close and we will reveal it soon (don't spoil it dumbass)". It could mean anything really.

I hope you have your popcorn. I found these:

http://vr-zone.com/articles/rumour-nvidia-geforce-gtx-680-to-release-in-february/14552.html
http://www.obr-hardware.com/2012/01/kepler-is-on-track.html

Both rumors saying practically the same. Launch in February. Still rumors, maybe true, maybe not...

I do know that if I was Nvidia and had a winner, I'd try to hide it until launch for a greater impact. Like they did with 8800 or like AMD did with HD48xx launch. Something like HD48xx/GTX2xx is not very likely to happen ever again, but I did say in another thread that IMO AMD pushing and pulling HD7970 and 7950 release dates around the way they did was a defense against that precisely. While only rumors, pretty much the same can be said about Kepler.

IMO this is what has happened lately:

1- AMD hears of Kepler (GK104) being at TSMC already, for production.
2- AMD pulls release date of HD7970 to December, to maximize sales at $550. HD7950 on Janary 9.
3- AMD gets more concrete info about Kepler. Specifically that it might be very competitive with Tahiti.
4- AMD delays HD7950 as a defense, no concrete release date, but February is the most likely option. Nvidia is going to talk about Kepler at CES according to non-confirmed schedule, so after knowing what they will be facing, a date and a price will be set for Tahiti Pro. Nvidi won't play a HD48xx on them.
5- Nvidia cancels any metion of Kepler at CES. When asked about why Huang only says to be patient.
6- Nvidia says they won't release Kepler until they have plenty of stock. The media estimates release date to be April.
7- Without any info on what exactly Kepler is, there's no benefit waiting anymore, so AMD pulls HD7950 release as much as they can, which happens to be the end of January.
8- (?) With the spy game pretty much over now, Nvidia pulls the release as much as they can, which is late February/March.

What you wrote is an interpretation with plausibility and an honest rendition of just how such volleys’ might be playing out.

Honestly, it was good to hear astute, knowledgeable, while nicely arranged timeline... good reading!

The only caveat in all this is what TSMC is providing in all this, as both are saddled to the same lame horse. It's like each of us reading the tea leaves.

So, if I hear what you're indicating is Nvidia will have/bring the GK104 that betters the 7970 and will be MSRP less than $550? Along with that a version that's betters the 7950 and less than $400? Finally, later there's still the GK110 (in maybe two derivatives) that move even higher than the 7970?

 

[purecain]

well i'm really happy to hear nvidia are coming back as usual... i want a 7970 cheap...bring on the price wars...

 

[EarthDog]

So it's ok for a company to crap out same performance cards each series?

No... you bought a midrange cards expecting it to beat out 2 high end cards (or dual GPU w/e). This is the way it always was. Next gen midrange is about equal to last gen high end. Naming conventions ftl!

That said, I hope we get seeded more than we did with the paper launch of the 7 series!

 

[Steevo]

FUD for ATI buyers.


Wood screws and drivers that kill cards.



Really I hope they get it out as it will provide competition for the consumer and make everyone happy.



Bene, it doesn't matter if "clock for clock" they are just as fast, as watt for watt, and in the real world we can buy a 1.3 Ghz 7970 from Sapphire, soundly kicking the shit out of the green offering. If this were a competition between Intel and AMD or if the tables were turned you would be saying that it doesn't matter.


Now, on with the thread.

 

[Benetanegia]

So, if I hear what you're indicating is Nvidia will have/bring the GK104 that betters the 7970 and will be MSRP less than $550? Along with that a version that's betters the 7950 and less than $400? Finally, later there's still the GK110 (in maybe two derivatives) that move even higher than the 7970?

I do think so, I think that a fast and relatively cheap GK104 is posible, based on what GF114 and GF110 were and the fact they have 50% more transistors to play with if they want to stay on the die size levels of Tahiti. So I honestly think the chances are very high.

But you never know. On price being "low" the only real evidence is that GK104 is said to be the replacement for GTX560 series, so a $500+ price tag is not very likely, at first. I cannot see Nvidia expecting to sell such a replacement card for more than $400, not even much more than $300, when the card that is replacing costs as low as $200 now. It's not realistic. But IMO AMD played too much on the safe side and fell short with Tahiti, underwhelmingly short IMHO. I know what people think and time will tell but that's my opinion. So in the end the price will be based on the reality of the market and we know that. If it's faster by any means, it will cost as much or almost as much.

There's GK110 tho, so I don't see a place for a $500 GK104 no matter it's performance, later they would need to also market 2-3 SKUs based on the big chip and the reality of the marketplace does not allow for $700 cards unless they are really really fast. i.e. dual-GPU kind of powerful. I don't see that happening either, and setting a high price now, say $500, when you know it's going to be replaced/raped in 6 months (or less than) by an GK110 SKU, and hence need to lower it's price to $250-ish, it's not wise, it would piss off far too many people.

So it's maybe time for chips that are somewhat smaller and cheaper IMO, because the big change already happened with Fermi and it's time to simply add some more GPCs or more SMs and extract the most out of it, and I think that this must have been Nvidia's plan, to an extent. But it's just my own opinion based on what I can see and what I don't see is Nvidia's purposes, so...

Bene, it doesn't matter if "clock for clock" they are just as fast, as watt for watt, and in the real world we can buy a 1.3 Ghz 7970 from Sapphire, soundly kicking the shit out of the green offering. If this were a competition between Intel and AMD or if the tables were turned you would be saying that it doesn't matter.

It does matter. TSMC talked about 28nm clocks being up to 45% faster, they have been talking about that for more than a year, and while maybe not to such an extent, higher clocks have been confirmed by pretty much every customer. Next Snapdragon runs faster, Krait runs faster, 28nm is faster, it's a fact or as much of a fact as it can be right now. The high clocks in Tahiti just only kinda demostrate it too. So you can't just assume that Kepler won't have 10-20% faster clocks. You really need to forget anything you think you know about AMD vs. Nvidia architectures, because Tahiti != Cayman, it's not VLIW, it's not super efficient anymore, it's not inherently and forcefully more Mhz friendly. Tahiti might overclock so well because of the arch change or because 28nm really brings some improvements, and everything has to be weigted in if you pretend to make an educated guess. It's going to be A, because it was A the last 2 years is not an argument, is only a falacy.

 

[Steevo]

Actually by my calculations the 7970 is about 5% faster clock for clock in almost every game.

Clock difference is 1.1981865 (925/772) 19.8% faster clocks

On most of the games the range in performance difference is

1.23-1.547 faster, subtract the difference 3-7% faster per clock, on average 5%.

And most importantly to gamers the difference at high resolution with eye candy the 7970 pulls ahead even more at 7% average faster per clock.

"You really need to forget anything you think you know about AMD vs. Nvidia architectures, because Tahiti != Cayman, it's not VLIW, it's not super efficient anymore, it's not inherently and forcefully more Mhz friendly. Tahiti might overclock so well because of the arch change or because 28nm really brings some improvements, and everything has to be weigted in if you pretend to make an educated guess. It's going to be A, because it was A the last 2 years is not an argument, is only a falacy. "



This makes no sense, I wasn't commenting on anything you just said, it is meaningless, I don't care if there are two babies fighting to the death in my graphics card to power it, as long as its fast. We were talking about the total performance figures for the card. 20Ghz or 2, as long as it is faster, and overclocking is just icing on the cake.

 

[Benetanegia]

Actually by my calculations the 7970 is about 5% faster clock for clock in almost every game.

Clock difference is 1.1981865 (925/772) 19.8% faster clocks

On most of the games the range in performance difference is

1.23-1.547 faster, subtract the difference 3-7% faster per clock, on average 5%.

And most importantly to gamers the difference at high resolution with eye candy the 7970 pulls ahead even more at 7% average faster per clock.

You are grasping at straws... I did not say it was the same, I said:

clock for clock they are mostly equal.

5% ?? 7% ?? That's minuscule and nonconsequential to my argument. Tahiti has 43% more transistors than GF100/110.

And when compared to GF104, with it's reduced GPGPU fetures, a trait that GK104 will most probably inherit, Tahiti is 220% the size.

I don't care...

Yeah, right, that's the problem. You rather overlook any evidence and just go around calling people biased, than really thinking about it and get hurt because of your own heavy bias.

 

[Steevo]

LOL wut?



I like you, your funny.

 

[EarthDog]

What is a bit comical to me is that one cant really compare clock speeds between the two architectures as it is so different so that entire point is based off of.....well nothing.

Watt for watt is also a pretty interesting discussion point as well. I mean think about it, who cares about watt /watt performance of a card that costs $550? If Im plunking down that much coin on a GPU, I likely just care about its raw performance and not how it compares watt /watt (or clock for clock for that matter) with another card. All I know is on games, regardless of clock speeds and overclocks and power consupmtion this card is a stud and wipes the floor with Nvidia (until Kepler).

 

[overclocking101]

now children just calm down, you are all fighting about who is faster. we dont even know yet!

 

[Casecutter]

Benetanegia - Your scenario, that a GK104 can better the 7970 and significantly undercut (20-25%) the $550 would be a boon for Nvidia, especially if they hit end February-March; although, I'd be astound if they get there. Likelier is that Uber AIB visions (like FTW editions with the GTX 460) of GK104 will jump a 7950, but consider 7950 are said to MSRP $400… it's all relative. With all that said, I think there's still plenty of room for AMD to be aggressive on price, so at this point wouldn’t persuading anyone to buy. Hold off... in this card game.

While, by the time GK110 hits I'd think as you said, AMD played too much on the safe side and fell short with Tahiti, underwhelmingly short IMHO". I might see that more based on production and ability to get good yield out of TSMC. While leaving clocks artificially low, and now permitting AIB's to build Uber versions (ie: like Nvidia) which many are releasing. By June I could see TSMC finally have their process honed-in (that’s being nice), and AMD on a re-spin of Tahiti, which garners achievements in clock speed and yields. Then we might see AMD add Rambus XDR2 (sure a huge aspiration, but adds variation), which might be enough in price savings to combat Nvidia. Such a move probably won’t take the crown, but really hit on the Bang-for-Buck maintaining at a $550 MSRP.

It right there in the "tea leaves" I notice...

Then on patience;
Romans 15:4-5

 

[Doperguy667]

yep

Was hoping for a little sooner but this gives me something to look forward to. Going to ride out my GTX470 till the new 28nm gpus dust settles.

me too waiting out with my old 470 as well

 

[EarthDog]

Yay OBR, hating on that tool Charlie... and providing a link to something more accurate than the other FUD going around!!

http://www.chiphell.com/thread-338350-1-1.html

 

[xenocide]

Yay OBR, hating on that tool Charlie... and providing a link to something more accurate than the other FUD going around!!

http://www.chiphell.com/thread-338350-1-1.html

How exactly is that more accurate?