GTX 680 Release Date?

[3DVagabond]

If GK100 is in fact in house and ready to be released in 4 months why call this card a GTX 680? Wouldn't it be better to humiliate AMD by releasing a card named GTX 660 Ti that beats a HD 7970 and then later on release a GTX 680 that completely roflstomps the HD 7970?

The only logical conclusion is that GK100 is not ready for at least another 5 months or more so nVidia decided to bump up the GK104 to a GTX 680 since it was competitive with a HD 7970.

Pure speculation so I can be completely wrong here.

They are in business to make money. Releasing it as the 680 means they can charge 680 prices for it. Releasing it as the 660ti means they can't charge as much for it. "Humiliating your opposition" is not what a company is in business for.

 

[Keysplayr]

Wow, I didn't know that HD3000 could do DX11 tessellation! :awe:

Maybe it was windowed so he could show GPU-Z and EVGA Preicison on the same screen? Dunno.. Do you by any chance happen to know the "real stuff" because of being a Focus Member? I know it's wise to try to keep people from being over-hyped and then therefore disappointed, but I'm just curious, man!

But it actually says "1680x900 Windowed" in the Unigine result screenie.. So that probably means it was run in windowed mode.
As far as "knowing". Can't say.

 

[ocre]

....they had to do it out of necessity because GK100 could not pass engineering sample status due to interconnect problems. Personally I find anyone suggesting that an "unreleased product" is really the high end, even though it won't be released until the end of the year, to be pretty funny. We can make a laundry list of unreleased products and proclaim them high end, such as haswell, sea islands, etc.

That is why GK104 is the flagship, for now.

Would you please share how could you possibly know this?

 

[Mopetar]

Would you please share how could you possibly know this?

Was something that made a run around the rumor mill a while back. If nVidia releases GK104 as the 680, it pretty much confirms that there was an issue with GK100 that lead to its cancellation.

 

[ocre]

Was something that made a run around the rumor mill a while back. If nVidia releases GK104 as the 680, it pretty much confirms that there was an issue with GK100 that lead to its cancellation.

there is a big difference between "an issue" and what he wrote.

 

[Keysplayr]

Was something that made a run around the rumor mill a while back. If nVidia releases GK104 as the 680, it pretty much confirms that there was an issue with GK100 that lead to its cancellation.

In reality, it would do no such thing.

 

[Mopetar]

In reality, it would do no such thing.

So you're saying that either nVidia decided not the make a GK100 part at all or that they made one but decided not the release it?

If GK104 is launched as the 680 it suggests that they won't have a bigger die out for a while. Given the company's recent naming policy, XX4 hasn't been their biggest die and the biggest die is usually GX0 or GZXX0.

Do you have some alternative theory that you'd care to present?

 

[RampantAndroid]

I'm a wee bit confused about this, the 104 in the 5xx series was mid range, now we have a 104 in the 6xx series and it's now a high end card?...does that mean no 110?, or it is a cheap shot by NV to get high end price with a mid range chip?

Or is this their way of releasing a GTX690 that'll have great perf and not be dual GPU?

Look, it beats the previous top end card. Who cares what it is numbered or called. The performance speaks for what it is. If they can hold back on tech they have now to incrementally release new chips fore more money, or to cause upgrades, and not do any R&D for it, they will. Welcome to capitalism.

 

[Grooveriding]

Or is this their way of releasing a GTX690 that'll have great perf and not be dual GPU?

Look, it beats the previous top end card. Who cares what it is numbered or called. The performance speaks for what it is. If they can hold back on tech they have now to incrementally release new chips fore more money, or to cause upgrades, and not do any R&D for it, they will. Welcome to capitalism.

I really doubt this. The GK104 practically screams to be used on a dual-gpu card. Low TDP, efficient, good thermals etc. This chip is ideal for a dual-gpu card and I have next to no doubt we will see a GTX690 with two of these on board.

If nvidia had a GK110 ready and a TSMC in a position to fab enough to field the product on shelves, we'd have it instead of GK104. GK104 is the best they have, GK104 is their flagship. $550, GTX680 name, marketing materials calling it 'the fastest DX11.1 GPU' What more can they do to convince anyone with doubt ?

My bets are GK110 will wind up being GTX 780, my hope is that it is no further off from GTX680 than the 580 was from the 480. The 680 is turning out to look like a shift in strategy from nvidia for their flagship. We've gone from big dies, wide bus width, more VRAM than seen previously offering 60% more performance than their previous flagship, to a little die, small bus, average VRAM on board - that based on possibly valid leaks is about 40% better at best. :thumbsdown:

It's like they've gone all AMD on us.

 

[RampantAndroid]

I really doubt this. The GK104 practically screams to be used on a dual-gpu card. Low TDP, efficient, good thermals etc. This chip is ideal for a dual-gpu card and I have next to no doubt we will see a GTX690 with two of these on board.

If nvidia had a GK110 ready and a TSMC in a position to fab enough to field the product on shelves, we'd have it instead of GK104. GK104 is the best they have, GK104 is their flagship. $550, GTX680 name, marketing materials calling it 'the fastest DX11.1 GPU' What more can they do to convince anyone with doubt ?

My bets are GK110 will wind up being GTX 780...

Sure, but then you're just applying the same logic, and making a different bet :sneaky:

That is to say, I believe the rumors that the GK110 exists, and may or may not be QUITE ready, and in large part because I think they realize that the 104 parts are sufficient to compete. Why show your deck of cards, when you need only reveal a single card that beats your opponent's card? Whether they make a dual GPU card or not is not the point I was making. Rather I was making the point that I suspect that Nvidia is holding back intentionally at this point. AMD releases the 79XX series. Nvidia releases the 104 in response. AMD releases something faster, and NV replies back with the 110 or something else.

 

[BallaTheFeared]

I don't have any answers... But uh...

Why sell a small die chip for $250 when you can sell it for $550 and increase your profit margin by over 100%?

Why release a big chip for $550, when you can sell your small one for $550?

Nvidia is looking at great profit margins with GK104, why bang out a much larger die on an expensive fab when your small die is enough to hold the performance crown?

If Nvidia can get $550 for GK104, why would they cannibalize their profit margins when they're just going to be competing with themselves from $250 to $550? Sure they could push AMD to the sub $250 market once again, but those markets are the best place to make money anyways... It wouldn't hurt AMD or help nVidia as much as the benefit of selling even less expensive to make gpus in that market will help both of them.

Companies win, consumers lose.

Just my opinions/thoughts on the subject.

 

[Mopetar]

Rather I was making the point that I suspect that Nvidia is holding back intentionally at this point.

Simpler logic suggests that nVidia doesn't have anything better at this point. You may as well just say that AMD is holding back too and realize how ridiculous that would sounds. Unless GK104 launches as something other than the 680, nVidia won't have anything better for the immediate future. eventually they'll come out with something bigger and better, and the wait may be as short as it was between the 480 and 580, but to suggest that they're holding back intentionally is plain silly. They're a business, not some kind of charity that only wants to release GPUs on par with AMDs high end.

 

[Mopetar]

I don't have any answers... But uh...

Why sell a small die chip for $250 when you can sell it for $550 and increase your profit margin by over 100%?

Why release a big chip for $550, when you can sell your small one for $550?

Nvidia is looking at great profit margins with GK104, why bang out a much larger die on an expensive fab when your small die is enough to hold the performance crown?

Because they can sell the big die for even more. If people will pay $550 for the smaller die, they pay more for a bigger die with better performance.

The rumors were that the big die had sme complications, so it was canceled. Fortunately nVidia's big die is much bigger than AMDs biggest die, so their next biggest die is fairly similar to AMDs high end, biggest die GPU. If you think nVidia didn't want to have a bigger die ready for to launch in this time frame you're just deluding yourself.

 

[BallaTheFeared]

I could be deluding myself, but I happen to already consider the $500+ market a fraction of their revenue stream. That could be part of the problem, the idea of Intels Extreme edition $1000 chips being their bread and butter never sat well with me either.

 

[ocre]

even if they could sell these for a lower price, there is a large supply constraint. Nvidia has tons of contracts they are struggling to fill, the majority moble. there is only so much TSMC can deliver. And everyon is fighting over the limited 28nm wafers.

Nvidia has a massive problem. They cannot keep up the supply and the discrete market will be effected. Prices will be high because of this anyway. Everyone wants their wafers but none of them realized how much they would overwhelm TSMCs ability to deliver.

 

[Mopetar]

I could be deluding myself, but I happen to already consider the $500+ market a fraction of their revenue stream. That could be part of the problem, the idea of Intels Extreme edition $1000 chips being their bread and butter never sat well with me either.

It's not their biggest market in terms of revenue, but how does having a $550 GK104 card and, hypothetically an $800 GK100 card stop nVidia from selling cards for $200 - $300?

Given that production is limited because both companies are using a new process, there's no reason not to aim for higher prices. Setting the prices lower just means you be sold out of cards for an extended period of time and people will piss and moan about there not being enough cards.

 

[SirPauly]

It's like a damned if you do or damned if you don't decision. Yields, costs and wafer constraints may be indeed issues but hopefully this chip may offer some solid performance/value -- so far its been more evolutionary or incremental.

 

[BallaTheFeared]

You nailed it though, neither Nvidia nor AMD have enough wafers... Why take away wafers from a smaller, higher quantity product that will produce higher sales volumes and thus more revenue simply to go "oohh yeah, pwned j00 AMD" with a silly $800 product that very few people would actually buy?

It seems to me wafers are the most valued commodity currently, so why go for bigK when you simply do not need it currently?

Leaks say Nvidia taped out GK100/110 several months after GK104. Perhaps their stratagy is to get the smaller easier to produce chip out quickly, then follow up with a refresh later on a cheaper, more refined 28nm process and design with more R&D and refinement?

You don't need the GTX 480 right away with all it's problems if you can release a GTX 460 first that beats the 5870, right?

 

[Grooveriding]

Finding the need to spin and explain the lack of GK110's presence as a choice rather than a reality of design and timeline pointless.

So GK107 is what, low end moved to mid-range ? What of GK106, guess that is the card nvidia planned to give away for free ?

No one knows squat. The only fact is GK104 is what nvidia is releasing as a high end GTX 680 for $500+. Let's wait and see if it is consistent with the performance increase nvidia usually delivers with their flagship on a new process.

 

[Mopetar]

You nailed it though, neither Nvidia nor AMD have enough wafers... Why take away wafers from a smaller, higher quantity product that will produce higher sales volumes and thus more revenue simply to go "oohh yeah, pwned j00 AMD" with a silly $800 product that very few people would actually buy?

Plenty of people would buy it and the highest end part can always have the highest margin based on the fact that it doesn't have anything to compete against. The only reason to release anything else is that there aren't enough people who buy at that level.

Given the limited production capacity you'd want to come out with the high end first. The people who buy those cards will pay for even marginal upgrades and you won't have sufficient production to supersaturate that market segment at launch.

Leaks say Nvidia taped out GK100/110 several months after GK104. Perhaps their stratagy is to get the smaller easier to produce chip out quickly, then follow up with a refresh later on a cheaper, more refined 28nm process and design with more R&D and refinement?

You don't need the GTX 480 right away with all it's problems if you can release a GTX 460 first that beats the 5870, right?

Historically, nVidia has always tried to release their big die first, because if production is going to be limited it's better to get as much profit as possible out of it to fund additional development. It's the reason why AMD launched the 7900s before anything else. There's more money to be made there.

I don't know if you were trying to make some historical reference there, but the 480 (GF100) launched before the 460 (GF104) by about two and a half months. The last time an X60 product launched first was the 260, but that actually based on the same GT200 die as the eventual 280, but with disabled cores.

In recent history, nVidia hasn't taken the approach you've outlined. They usually try to get their big die out first, and if they they run into the problems they just bin the parts or make sure that the eventual refresh solves any issues. Furthermore, pretty much any existing rumor points out that they attempted this same approach again, but for whatever reason decided to cancel their biggest die.

Your hypothesis does not make sense from either a financial or historical perspective, and furthermore existing rumors seem to go against it as well. I can't see any reason to continue assuming that line of reasoning outside of better support for your arguments.

 

[bunnyfubbles]

Looks like Kepler will indeed support DX11.1

support for the next version of DX has never been a legitimate advantage, the GPUs end up being simply too slow by the time the respective technologies are viable

we're still slowing growing into DX11, and I certainly wouldn't want to run any serious DX11 title with my now relatively archaic 5850 even though it does officially support DX11.

whats more is that these .1 versions also never really pan out to be a true advantage - it never amounted to anything for ATI/AMD with their 3000/4000 series, to this day there's simply no way I'd take a 4890 (DX10.1) over a GTX285 (DX10.0) if I was forced to use one or the other.

Same goes for DX8.1 with the Radeon 8500 vs. the DX8 GeForce 3

 

[SpicyTime]

Plenty of people would buy it and the highest end part can always have the highest margin based on the fact that it doesn't have anything to compete against. The only reason to release anything else is that there aren't enough people who buy at that level.

Given the limited production capacity you'd want to come out with the high end first. The people who buy those cards will pay for even marginal upgrades and you won't have sufficient production to supersaturate that market segment at launch.

Historically, nVidia has always tried to release their big die first, because if production is going to be limited it's better to get as much profit as possible out of it to fund additional development. It's the reason why AMD launched the 7900s before anything else. There's more money to be made there.

I don't know if you were trying to make some historical reference there, but the 480 (GF100) launched before the 460 (GF104) by about two and a half months. The last time an X60 product launched first was the 260, but that actually based on the same GT200 die as the eventual 280, but with disabled cores.

In recent history, nVidia hasn't taken the approach you've outlined. They usually try to get their big die out first, and if they they run into the problems they just bin the parts or make sure that the eventual refresh solves any issues. Furthermore, pretty much any existing rumor points out that they attempted this same approach again, but for whatever reason decided to cancel their biggest die.

Your hypothesis does not make sense from either a financial or historical perspective, and furthermore existing rumors seem to go against it as well. I can't see any reason to continue assuming that line of reasoning outside of better support for your arguments.

Well said. It takes more than a year to design and deliver a GPU, especially one on a new process. Nvidia read the 7970 benchmarks in December 2011, which was 2 or 3 months before their rumored GK100 release. Two to three months is not enough time to shift gears and intentionally delay GK100, then bump GK104 to the flagship in an attempt to make more money. If you think Nvidia is playing poker, holding onto a winning hand and waiting for the right time to use it, you're delusional. This is the tech industry. You get your best product out as fast as possible, then work on the next iteration.

Just my speculation as well, but I'm thinking Nvidia went super ambitious and aimed for GK100 to destroy the 7970 at 500mm2, and aimed for GK104 to destroy the 7870 at 300mm2. The GK100 couldn't yield, so plan B is to send GK104 up against Tahiti.

 

[AtenRa]

I dont believe NV will use the GK104 for the high end GPGPU/HPC market. We could see the GK100 in Q2 for the Desktop and Q3 for the HPC segment.

 

[Cookie Monster]

For those thinking nVIDIA cancelled GK100/110 or whatever it is, its simply not viable at all unless your intel and even then it would be questionable. The amount of investment in terms of money and time simply does note equate to "oh, it will not make it in time, lets pull the plug". Plus the whole "time" thing I find hilarious. Its as if they are working under AMD's time schedule..

 

[Silverforce11]

Cancelled? No, R&D is too costly and its time-frame would ruin everything follow up from it.

Delayed? Yep.

Hopefully gk104 will be as amazing as these leaks have it, its enough to hold on to the high-end for awhile.