Kepler Nvidia GeForce GTX 780

[DilTech]

I'm not totally sold that there is some super secret big gk110 around the corner waiting to pounce, there might be I don't know, but when I think of Nvidia gk104 I get the feeling it is so not Nvidia for a high end product.

Nvidia has always been the go big or go home type brand, gk104 specs seem so much more mainstream palatable across the board than their normal over the top feeling high end offerings.

If they do have a big gk core waiting I would imagine they'll be holding off until AMD's refresh and push gross margin with the gk104 in the mean time.

If gk104 can deliver the needed performance to command a ~$500 price tag, whether it was intended or not, it doesn't make allot of sense to sell a higher cost product at the same price point if not absolutely necessary to be competitive.

As far as I'm concerned if Nvidia is releasing gk104 as a high end product then it is in fact their high end product and named as such. Anything else is going to have to be next gen next release hardware since they've already allocated their high end 680 labeling for the gk104 and will likely be offering a dual card based on gk104. That in my mind leaves little chance or logic of offering a separate gk110 big die uber performance part until 7xx series is up to bat.

We'll see...

People figure there's another part sitting and waiting for multiple reasons.

1.) It's strange for NVidia to release a new architecture's high end part under a code-name ending in any number but 0. Refreshes are a different story, but this is clearly not just a refresh.

2.) NVidia haven't made a small chip high-end in quite some time, not counting refreshes.

3.) NVidia's statement that they were actually disappointed and surprised by the lack of performance in AMD's 7970. That kind of statement makes you figure that NVidia had a much faster chip ready to go, saw they didn't need it, and set it aside.

Seriously, all signs clearly point to NVidia having something else but see the opportunity to make bank off what would have been their mid-range part.

[Dalten]

I still want to know, how do the rumored 1536 shaders and 128 texture units come from only 500m~ increase in transistors?

[kuroikenshi]

Sitting on your cards for too long leaves them smelling like crap. Blowing out old stock is fine. Rolling out new parts is fine, but dont do it too long or you might end up with a case of self-inflicted butt-hurt.

I'm glad none of you guys are handling my money.

with the way they talk about "sound business decisions" its fair to say most who put forth this sort of plethora ARE basing their analogies on nothing more but sheer lack of business management experience and first and foremost, the lack of "ACTUAL INFORMATION"

[DilTech]

I still want to know, how do the rumored 1536 shaders and 128 texture units come from only 500m~ increase in transistors?

You also have to factor in the removal of 128-bits on the memory bus.

[highoctane]

Why couldn't Nvidia just call the GK110 a 690 and the dual GK104 a 685 and dual GK110 695? The naming structure isn't holy and unchangeable...

While the general naming scheme can be changed to whatever it doesn't change very often with Nvidia. If there was some hint that the naming was being refreshed to something new I could understand but its looking to follow the same hierarchy used since 2xx series.

Not only that but I doubt there will only be one video card model based on the possible gk110 gpu which leaves little room in the current name scheme, at minimum there should be two models offered due to binning & harvesting gpus.

[highoctane]

You also have to factor in the removal of 128-bits on the memory bus.

I'm guessing there has to be major changes to the cores themselves. If the reported transistor count and die size is correct its the equivalent of gpu gastric bypass surgery, they will have managed to get ALLOT more from much less.

[Dumo]

GTX 680 SLI from VR-Zone...http://www.microsofttranslator.com/b...11-03182012%2F

[cold2010]

VR-Zone

[The Jesus]

GTX 680 SLI from VR-Zone...http://www.microsofttranslator.com/b...11-03182012%2F

Are these cards not PCI-E 3.0?

[zerazax]

People figure there's another part sitting and waiting for multiple reasons.

1.) It's strange for NVidia to release a new architecture's high end part under a code-name ending in any number but 0. Refreshes are a different story, but this is clearly not just a refresh.

2.) NVidia haven't made a small chip high-end in quite some time, not counting refreshes.

3.) NVidia's statement that they were actually disappointed and surprised by the lack of performance in AMD's 7970. That kind of statement makes you figure that NVidia had a much faster chip ready to go, saw they didn't need it, and set it aside.

Seriously, all signs clearly point to NVidia having something else but see the opportunity to make bank off what would have been their mid-range part.

Regarding point 3, that's also marketing speak and PR. They've been doing that for ages

As for point 1 and 2... well, same could've been said about AMD when they released the 4870 and then the 5870. It was strange for AMD to release a non-refresh part under the RVx70 code, yet they were stunning successes.

Nvidia having something big in the works is certainly known, but consider two other important points:

1) AMD used smaller chips first when changing processes, e.g. 55nm to 40nm transition. It saved them from a Fermi situation. Nvidia might have learned that lesson

2) The GK110 or whatever you want to call it simply isn't ready yet, and won't be til the end of the year

There's far far far too much speculation about a mysterious -110 chip with little to no hard evidence on it which is why it is absurdly silly at this point to call anything a failure one way or another (cept the pricing it seems)

[Cold Fussion]

People figure there's another part sitting and waiting for multiple reasons.

1.) It's strange for NVidia to release a new architecture's high end part under a code-name ending in any number but 0. Refreshes are a different story, but this is clearly not just a refresh.

2.) NVidia haven't made a small chip high-end in quite some time, not counting refreshes.

3.) NVidia's statement that they were actually disappointed and surprised by the lack of performance in AMD's 7970. That kind of statement makes you figure that NVidia had a much faster chip ready to go, saw they didn't need it, and set it aside.

Seriously, all signs clearly point to NVidia having something else but see the opportunity to make bank off what would have been their mid-range part.

Nvidia spends hundreds of millions if not billions designing these chips, it's in their best interest to see a return on that investment as soon as possible. The longer they wait to release it, the less it is worth and if they wait to long it may become worthless. The only logical reason why they wouldn't release it if they had is because it isn't ready for what ever reason.

[Russian]

Sitting on your cards for too long leaves them smelling like crap. Blowing out old stock is fine. Rolling out new parts is fine, but dont do it too long or you might end up with a case of self-inflicted butt-hurt.

I'm glad none of you guys are handling my money.

Lmfao. Quotable material for sure.

I'm no economist (not yet, atleast) but I'm fairly sure I understand NVIDIA product strategy and it makes a lot of sense considering what I've seen.

[boxleitnerb]

If you can only make a relatively little amount of GK110 chips, what do you do?

a) sell them on $500-600 Geforce cards, demand is high, availability is bad
b) sell them on $5000+ Tesla cards, meeting your contracts with HPC clusters

Holding back is silly, no argument there. Luckily for Nvidia, these chips can cater more than one market.

[STEvil]

If you can only make a relatively little amount of GK110 chips, what do you do?

a) sell them on $500-600 Geforce cards, demand is high, availability is bad
b) sell them on $5000+ Tesla cards, meeting your contracts with HPC clusters

Holding back is silly, no argument there. Luckily for Nvidia, these chips can cater more than one market.

c: proceed as always: sell some as high end consumer GPU's for gaming, sell others for HPC/Render/etc.

[boxleitnerb]

But it isn't as always. Nvidias beefed up "GTX660 Ti" of this generation (the GK104) is sufficient to be competitive. Would GF100 had come first (or at all for the desktop) if even GF104 could have beat the 5870? Apparently, there is no need to feed the gaming market with faster cards than the 680 right now because you would only be setting the bar higher for yourself and lowering prices (or selling this new GK110 card for $800, then good luck with that).

I hope/believe there is a GK110 card no later than September this year, though. But the problem is: Why should they release that if GK104 fares so well and AMD doesn't have a faster part? Rather wait until it is really necessary and doesn't cannibalize your other products.

[kuroikenshi]

But it isn't as always. Nvidias beefed up "GTX660 Ti" of this generation (the GK104) is sufficient to be competitive. Would GF100 had come first (or at all for the desktop) if even GF104 could have beat the 5870? Apparently, there is no need to feed the gaming market with faster cards than the 680 right now because you would only be setting the bar higher for yourself and lowering prices (or selling this new GK110 card for $800, then good luck with that).

That's like saying "why would Intel need Haswell or IB???" SB-E is competitive enough, and surely there is enough stock floating around, even from 775 socket.

The point is that end user train of thought has nothing to do with a company's directive objectives. Every video card the competition sells is profit they have lost on, its that simple.

[boxleitnerb]

Intels product cycle is a bit longer than a year now. Of course you release new products, but not 5-6 months after you launched the previous one. Look at how long Nvidia milked G80 and its derivatives without bringing a card that was faster than the 8800GTX/Ultra.

What exactly would Nvidia have to gain if they sold the few GK110 chips they can make in summer for 10% of the profit they could make with Quadros? GK104 will be very competitive if not superior to Tahiti, so it's enough to hold down the fort for the foreseeable future. The situation is ideal: Sell the small performance chip for highend prices and the highend chip for insane prices in the professional market.

[STEvil]

Selling the high end chip only in the professional market leaves you open to a blow from the competition, shows the competition what your high end could be (fatal move if they have performance optimization only a respin away, ala 4870/4890, or find that they can create a multi-chip card in the power budget that can beat the new high end card), and weakens your public image because it puts your "high end" equipment so far out of their reach that they will begin to view you as "the man" rather than a competative force or an underdog.

The fact is that nV will either put out a high end consumer part based on "110", or it will be vaporware and they will look like bitboys to some people... that or 3dfx is eating them up from the inside out.

[DilTech]

Nvidia spends hundreds of millions if not billions designing these chips, it's in their best interest to see a return on that investment as soon as possible. The longer they wait to release it, the less it is worth and if they wait to long it may become worthless. The only logical reason why they wouldn't release it if they had is because it isn't ready for what ever reason.

Considering right now they have the opportunity to make the amount they'd have charged for their GK110 on the MUCH cheaper to produce GK104 (cheaper die, easier to produce, and cheaper pcb as well), and still release the GK110 at the same price later, what's the problem on their end?

Yes, it SUCKS for us consumers--but really, no one with any actual business sense can fault them. AMD undershot this round; NVidia saw a chance to capitalized financially and they are pouncing on it. Now rather than selling one of their designs for this generation at high end prices, they're going to be able to get away with doing so with two of them.

I still find it funny how many people are in denial about this whole thing. Think about it for a minute guys...

Last year, AMD 7970 still rumors in terms of performance and specs, we catch word of GK110 and GK104. It stayed this way until we saw the numbers out of the 7970, suddenly it just gets dropped to GK104? Combine that with NVidia's statement about being disappointed by the performance of the 7970. Now add in my previous statements... Only the most extremely naive people would believe that NVidia intended on starting out with the GK104 as the high end. They're doing it because they can. When there's a lack of competition, the consumers are the ones who suffer. Right now AMD can't compete with NVidia in that space, and as such we are the ones to suffer for it.

The next question though, if GK104 got moved to the high end space, what is going to take it's place in the mid-range?

[boxleitnerb]

Interesting point there. The question is, could AMD react? I expect GK110 to be at least 40% faster than GK104. According to some it will be a bit above 500mm2 and pack 6bn transistors. Bandwidth would also be considerably higher. It would be a bit like GF114 vs GF110.

But I admit that it takes time to get a product ready for launch, even if the chip is ready. Maybe Nvidia wants to be first with this "refresh". However, they could just clock GK104 higher, they have more headroom in regard to TDP.

[spicypixel]

I figure GK107 is taking the midrange space.

[STEvil]

Right now AMD can't compete with NVidia in that space, and as such we are the ones to suffer for it.

Who says they cant? You cant say anything until all the cards are on the table.

[Russian]

Who says they cant? You cant say anything until all the cards are on the table.

Then we all better keep our mouths shut for quite some time.

[DilTech]

Interesting point there. The question is, could AMD react? I expect GK110 to be at least 40% faster than GK104. According to some it will be a bit above 500mm2 and pack 6bn transistors. Bandwidth would also be considerably higher. It would be a bit like GF114 vs GF110.

But I admit that it takes time to get a product ready for launch, even if the chip is ready. Maybe Nvidia wants to be first with this "refresh". However, they could just clock GK104 higher, they have more headroom in regard to TDP.

AMD will have a few options. A dual 7970 is definitely a possibility, and I'm sure a few companies will put out dual 7870's (which could end up the bang for buck champ if priced properly). It's also possible that AMD are sitting on another design and planning a quick launch that'll normalize pricing (think x1800-x1900), although that situation is doubtful since all we've seen is a slide that was proven to be fake.

[Russian]

AMD will have a few options. A dual 7970 is definitely a possibility, and I'm sure a few companies will put out dual 7870's (which could end up the bang for buck champ if priced properly). It's also possible that AMD are sitting on another design and planning a quick launch that'll normalize pricing (think x1800-x1900), although that situation is doubtful since all we've seen is a slide that was proven to be fake.

Let's be honest... right now, AMD's biggest limitation in putting out a dual-GPU card is that they'd have to significantly dial down the clock speeds in order to hit 300w. Their biggest enemies are heat and power consumption, sounds really familiar actually. lol