Wednesday, October 17th 2012
NVIDIA Kepler Refresh GPU Family Detailed
A 3DCenter.org report shed light on what NVIDIA's GPU lineup for 2013 could look like. According to the report, NVIDIA's next-generation GPUs could follow a similar path to previous-generation "Fermi Refresh" (GF11x), which turned the performance-per-Watt equation around back in favor of NVIDIA, even though the company's current GeForce Kepler has an established energy-efficiency lead. The "Kepler Refresh" family of GPUs (GK11x), according to the report, could see significant increases in cost-performance, with a bit of clever re-shuffling of the GPU lineup.
NVIDIA's GK104 GPU exceeded performance expectations, which allowed it to drive this generation's flagship single-GPU graphics card for NVIDIA, the GTX 680, giving the company time to perfect the most upscaled chip of this generation, and for its foundry partners to refine its 28 nm manufacturing process. When it's time for Kepler Refresh to go to office, TSMC will have refined its process enough for mass-production of GK110, a 7.1 billion transistor chip on which NVIDIA's low-volume Tesla K20 GPU compute accelerator is currently based.
The GK110 will take back the reins of powering NVIDIA's flagship single-GPU product, the GeForce GTX 780. This product could offer a massive 40-55% performance increase over GeForce GTX 680, with a price ranging anywhere between US $499 and $599. The same chip could even power the second fastest single-GPU SKU, the GTX 770. The GK110 physically packs 2880 CUDA cores, and a 384-bit wide GDDR5 memory interface.
Moving on, the real successor to the GK104, the GK114, could form the foundation for high-performance SKUs such as the GTX 760 Ti and 760. The chip has the same exact specifications as the GK104, leaving NVIDIA to tinker with clock speeds to increase performance. The GK114 will be relegated to performance-segment SKUs from the high-end segment it currently powers, and so even with minimal increases in clock speed, the chip will have achieved sizable performance gains over current GTX 660 Ti and GTX 660.
Lastly, the GK106 could see a refresh to GK116, too, retaining specifications and leaving room for clock speed increases, much in the same way as GK114, except, it gets a demotion to GTX 750 Ti, GTX 750, as well, and so with minimal R&D, the GTX 750 series gains a sizable performance gain over its previous generation.
Source:
3DCenter.org
NVIDIA's GK104 GPU exceeded performance expectations, which allowed it to drive this generation's flagship single-GPU graphics card for NVIDIA, the GTX 680, giving the company time to perfect the most upscaled chip of this generation, and for its foundry partners to refine its 28 nm manufacturing process. When it's time for Kepler Refresh to go to office, TSMC will have refined its process enough for mass-production of GK110, a 7.1 billion transistor chip on which NVIDIA's low-volume Tesla K20 GPU compute accelerator is currently based.
The GK110 will take back the reins of powering NVIDIA's flagship single-GPU product, the GeForce GTX 780. This product could offer a massive 40-55% performance increase over GeForce GTX 680, with a price ranging anywhere between US $499 and $599. The same chip could even power the second fastest single-GPU SKU, the GTX 770. The GK110 physically packs 2880 CUDA cores, and a 384-bit wide GDDR5 memory interface.
Moving on, the real successor to the GK104, the GK114, could form the foundation for high-performance SKUs such as the GTX 760 Ti and 760. The chip has the same exact specifications as the GK104, leaving NVIDIA to tinker with clock speeds to increase performance. The GK114 will be relegated to performance-segment SKUs from the high-end segment it currently powers, and so even with minimal increases in clock speed, the chip will have achieved sizable performance gains over current GTX 660 Ti and GTX 660.
Lastly, the GK106 could see a refresh to GK116, too, retaining specifications and leaving room for clock speed increases, much in the same way as GK114, except, it gets a demotion to GTX 750 Ti, GTX 750, as well, and so with minimal R&D, the GTX 750 series gains a sizable performance gain over its previous generation.
127 Comments on NVIDIA Kepler Refresh GPU Family Detailed
Unless I'm seriously misunderstanding you, you're arguing that GK110 is/was scrapped due to inherent problems with die size even though we're sitting here reading and commenting on an article that is proposing that GK110 is going to be released just fine for the 7xx series cards.
That makes no sense.
Much more likely is that their yields sucked last year on this chip and so they bumped GK104 up a tier from 660ti to 680, while putting GK110 on the back burner until they got yield issues fixed.
:eek:
:shadedshu
What's in the news doesn't make sense. period. I'm not gonna argue it's bogus...if you don't realize that yourself..well...I won't argue with you.
:D
It's twice the size of GTX680, this news says, but only 55% faster.
We know it'll use the smae silicon process as current GTX680...how can die size NTO be a problem?
If it wasn't, then a doubling of the same transitors should equal a doubling of performance, no?
semiaccurate.com/2012/10/15/will-nvidia-make-a-consumer-gk110-card/
Tahiti ≈ 365 mm²
Gk104 ≈ 295 mm²
Difference in size ≈ 24%
Difference in # of. transistors ≈ 33%
looking at the above numbers, Tahiti does pack more transistors but the die sizes are not too close either
However, there's nothing that inherently makes that the case. S|A's article (linked above) offers a fairly good explanation here, I think.
And of course, we're only dealing in rumorville, so we'll see what happens.
see, to me, a mid-range chip is under 250mm², like GTX 660 and HD 7870. All these claims of GTX680 being mid-range do not make sense.No, really, the smoking gun is that design schedules ALWAYS work this way.
See, nVidia and AMD are both contrained by what TSMC offers. They both buy wafers from TSMC, TSMC makes all chips for both, and as such, they are even using the same process.
AMD packs 33% more transistors into HD 7970. It's not 33% bigger.
Nvidia may be able to further increase transistor density, for sure, but it's not going to be enough that even qualifies GTX 680 as "mid-range"
I'm not denying it might have been planned...but reality says, since chips take liek 2 years to design, that they knew since day one it wasn't going to happen. They knew LONG before those "claims" came out that GTX 680 was the chip we got.
GK110 or GK100 or whatever...was NEVER meant to be GTX 680. Nor was it meant to compete with the current HD7970.
I'm not denying that a new GPU is coming, either. :p
Basically, I'm arguing that GK104 was not drawn up as a high end GPU. It wound up filling that role just fine, but that doesn't mean it was planned that way.
Because companies never do that...
anyways with due respect I wish to end it here, to each his own (its all speculation)..
All Nvidia high-end chips (GPU + HPC) of past generations have been close to or bigger than 500 mm^2. G80 484mm^2, GT200 576mm^2. GF100 520 mm^2.
Time to have a reality check man. GK100/110 IS the high-end chip. A chip that Nvidia decided it was not economically feasible this past months when TSMC supply was so constrained and yields (for everybody) were not good. End of story. It really is. There's no problem with it other than that and the fact that by being bigger it's going to have lower yields and lower number of dies, nothing that Nvidia didn't do previously or that are afraid of. GK106 took long to release too. Was it because it was not posible? No, because it was economically less "interesting" than GK104 and so was GK110. If they could win with a 294mm^2 chip there was absolutely no reason to release the big one and have lower margins as they had to with first Fermi "generation". HPC moves slower and relies on designs like the Oak Ridge supercomputer that would not been ready back at the time, so more reason to delay.
Of course bigger processes took up more space. :p
Silly.:roll:
I never said GK100 or GK110 is NOT the high-end chip...sure is...but it was NEVER meant to be GTX680.
TSMC had yield issues. :p That is comical. Yeah, blame the infant technology. :laugh:
Of course it was horrible. nVidia KNEW it would be, as did AMD...and they dealt with it, as they have with every process.Actually, no, i think nVidia did NOT make a big mistake, at all, and that GK100 was planned for next year ALWAYS, rather than this January or whatever.
It's not like Kepler is some new thing..it's a tweaked Fermi. nVidia admitted big mistakes with Fermi, so I do expect that there were extra-cautious with kepler.
As will be the next chip.
Except, of course, as a reviewer, I do have a bit more info than the average joe, although, not as much as many other reviewers do, I'm sure.
See, the difference between me and other reviewers..I do this for fun, as a hobby..and not for cash.
I'm not posting news for hits, because that garners money for the site with ads...
TPU isn't built upon that, at all.
This is specualtion, after all, not fact, so yeah, I offer a different perspective...So?
At the end of the day, it's me playing with the hardware NOW you guys want to buy IN THE FUTURE. I don't really care who has the faster chips, who is cheaper, or what you buy...this stuff just shows up on my doorstep, with ZERO cost.
I'm just not afraid to be wrong. :laugh: In the future, we can say "look, this was right, and this wasn't"...and I won't care if I'm wrong. :p You might...but I won't.:roll:
That's where me being a reviewer is important.
Who cares that I review stuff. It's not important, really. Like, really...big deal..I get to play with broken stuff 9/10 times, when it's pre-release. I've said it before, I'd much rather have stuff later, but I guess some OEMs value my feedback prior to launch. That's like the whole "ES is better for OC" BS.
That fact I do that for them, for free...well...it's not a big of a deal that most seem to think it is. I actually think it's kind of the opposite...
At the same time though, those that DO have info about unreleased products, like myself, also cannot say much, except what they are allowed, or their info cannot be real.
THAT is a fact I learned as as reviewer, that many seem to not know. That is just how it works. Either this info is force-fed, or it's fake.
No one's blaming the "infant tech". Both AMD and Nvidia design their chips according to TSMC's guidances on the process. They have to, since they have to design the chips long before TSMC is ready for production. They design around them and weight in the feasibility and profitability based on them. Guidances are one thing and reality is often a very different one. Of course AMD by being a fabbed chip maker in the past, knows better than Nvidia how to deal with them. We are not discussing that so to the point. Trying to deny that volume and yield issues are TSMC's problem is stupid. Ther guidances for the process and reality didn't match and everyone has suffered from it, be it Nvdia, Qualcomm or AMD, even if AMD has not been as vocal. Each company has very different things to address in their conference calls and trying to extract any conclusions from whether they talk about TSMC issues or not is again stupid. AMD is in far more trouble and has much more things to excuse than having to explain why profit margins on the GPU bussiness are slightly lower than expected.
So imagine we are Nvidia. 28nm is not as good as it was "promised" to be. We get close to Kepler release dates. Volume is not good, yields are not good either, neither worse then 40nm, as Jen Hsun Huang said. But Nvidia had 2 options, repeat GF100 or release GK104 as the high-end. The answer is simple. In a waffer you can have 201 GK104 die candidates. And ~100 GK110 candidates. Again, knowing that GK104 would be close to Tahiti performance or beat it, it's an easy choice*. GK104 at $500. There was no price point at which GK110 would have been more profitable, no matter how much faster than HD7970 it could have been. With the severely low 28nm volume, they would never be able to sell enough GK110 cards so as to be more profitable than they have been with GK104, even if they had acheved 100% market share.
* More so when you know that the next node willl not be ready until 2-3 years later. You'll have to do a refresh and you'll have to make it appealing, faster, so by doing what they did, they can kill 2 birds with a single stone.
en.wikipedia.org/wiki/Argument_from_authority#Disagreement
en.wikipedia.org/wiki/Appeal_to_accomplishment