Nvidia Kepler (6XX-series) discussion

[kidsafe]

I would not expect 20nm half-node from GlobalFoundries or TSMC until 2H 2013.

[Zeara]

http://vr-zone.com/articles/report-n...-ti/14952.html

Nothing special, if the performance part is true. But it will depend on the pricing ofcourse

[Rennadrel]

Which is quite ironic, seeing in AMDs other field, they are starting to become years after intel. Which is a sad thing.
A good thing they've realised, though, and shifts their interests from high-end CPUs and toward a stronger integrated GPU though. A shame for workstations and enthusiasts, but it's what will likely win them big shares of the market in the long run.

However, I think they should still have an architectual fine-tuning of 28nm(?) before they shrink again.

I think AMD's next generation will probably be a power requirement reduction rather then a pure die shrink. And honestly I can't see a die shrink being possible without a couple of years worth of development. But the top end of graphics technology right now is so far ahead of the game development industry that it seems pointless to do much until games start really utilizing top end technology.

It's kind of amusing and a shame that Nvidia is going to be so far behind AMD though, competition always pushes each business to be better, however if AMD winds up with such a substantial lead just purely based on time, Nvidia is going to have to put out a flagship card that absolutely steam rolls AMD's 7970 and 7990 cards.

[DeltrusDisc]

http://vr-zone.com/articles/report-n...-ti/14952.html

Nothing special, if the performance part is true. But it will depend on the pricing ofcourse

Sincerely hope this is true, been aching to hear something good from NVidia.

[Zeara]

Sincerely hope this is true, been aching to hear something good from NVidia.

Only good thing will be that they will launch something. Performance wise it isnt what ppl were expecting, or the name is wrong and it will be called a 660 in the end. Who knows :P

The 680 is still months away tho

[Butler Log]

The question I have is how small they can make those components. At some point they will reach the limits of what silicon, gold and silver can do - make a gold strand thin enough (not that much smaller than 20 nm) and it will no longer conduct electricity. For that reason I doubt we will be seeing any major reductions in die size for quite a while.

[DeltrusDisc]

The question I have is how small they can make those components. At some point they will reach the limits of what silicon, gold and silver can do - make a gold strand thin enough (not that much smaller than 20 nm) and it will no longer conduct electricity. For that reason I doubt we will be seeing any major reductions in die size for quite a while.

Well, Intel is soon releasing Ivy Bridge, 22nm die CPUs. ^_^ Should be interesting! And while you have a point, I'm sure the companies will figure something out. ;p

@Zeara, yeah but I would probably not opt for a 680, now a 670 Ti? I just love the Ti, having the 560 Ti myself, gotta get me some Ti, you know what I mean? TI.

[kidsafe]

It's kind of amusing and a shame that Nvidia is going to be so far behind AMD though, competition always pushes each business to be better, however if AMD winds up with such a substantial lead just purely based on time, Nvidia is going to have to put out a flagship card that absolutely steam rolls AMD's 7970 and 7990 cards.

Halo product tiers like the fastest video cards only improve Nvidia's overall revenues/margins/profits by single digit percentages. Nvidia isn't required to panic at all here...They just need to redress the current mobile Fermi line-up, add mobile Kepler and ship GK104/106 sometime this quarter. GK110 being delayed affects us consumers more than it does Nvidia.

[Romandix]

The question I have is how small they can make those components. At some point they will reach the limits of what silicon, gold and silver can do - make a gold strand thin enough (not that much smaller than 20 nm) and it will no longer conduct electricity. For that reason I doubt we will be seeing any major reductions in die size for quite a while.

I think they will move onto more economical materials better suited for the job and flip the fabrication process on its head sooner than we think possible.

[kidsafe]

I think they will move onto more economical materials better suited for the job and flip the fabrication process on its head sooner than we think possible.

We'll be die shrinking for years to come. I honestly think we'll see 5nm transistors before we have to switch to breakthroughs such as nanoscale carbon-based or optical technology.

[lvlark]

We'll be die shrinking for years to come. I honestly think we'll see 5nm transistors before we have to switch to breakthroughs such as nanoscale carbon-based or optical technology.

Optical? I always thought light has bigger wavelengths than electrons?

[tetrisGOAT]

Halo product tiers like the fastest video cards only improve Nvidia's overall revenues/margins/profits by single digit percentages. Nvidia isn't required to panic at all here...They just need to redress the current mobile Fermi line-up, add mobile Kepler and ship GK104/106 sometime this quarter. GK110 being delayed affects us consumers more than it does Nvidia.

GK110 haven't even been announced yet?

And I'm surprised to see anything about Kepler being released before June (although I read the article a good nine hours ago, I doubt it will be availible in stores in March).

[kidsafe]

Optical? I always thought light has bigger wavelengths than electrons?

I'm referring to the use of light to manipulate electrons rather than current.

[Rennadrel]

I think they will move onto more economical materials better suited for the job and flip the fabrication process on its head sooner than we think possible.

Being economical is not something Nvidia has been prone to doing in recent years. I think being economical has been and will continue to be up AMD's alley for the time being, even if Nvidia continues to put out the stronger cards, AMD will be more sensible for those who want a more environment friendly graphics card, which is almost ironic since AMD's CPU's are very high energy use, yet they try to go for less wattage with more performance every generation.

[tetrisGOAT]

Being economical is not something Nvidia has been prone to doing in recent years. I think being economical has been and will continue to be up AMD's alley for the time being, even if Nvidia continues to put out the stronger cards, AMD will be more sensible for those who want a more environment friendly graphics card, which is almost ironic since AMD's CPU's are very high energy use, yet they try to go for less wattage with more performance every generation.

To be fair, the AMD CPUs in general don't necessarily draw more power, per se.
They count TDP-values differently than intel does.
As far as Bulldozer goes, it wasn't that powerconsuming, unless overclocked, which is nothing they expect the general user to do.
The Sandy Bridge-E series also quickly ran away with power consumption, but had a bit more performance and performance gain to back it up, making it a worthwile thing.

[Lowlie]

Pure speculation:

GK100, alleged flagship: cancelled in its current form. (GTX 480 and 580 code names: GF100 and GF 110)

GK104 will be branded Geforce GTX 680. (Probably originally planned as the 660) It will perform similar to the 7970 and not really be what they had planned to put out as the top-segment product but GK100 might not have turned out well and AMD didn't set the 28nm bar very high anyway.

GK100's second attempt will be renamed GK110 and will get the retail name GTX 780. It could come out much later this year, closer to Q4 or even Q1 2013. The 600 series will have been a couple of re-brands of fermi, the GK104 and a few laptop chips, nothing more.

Wait half a year then tell me I'm wrong :P

I think you are exactly right with this. Im speculating the GTX680 to be around 3 to 6 % faster than the 7970HD (with no oc) so nothing special. But dont expect the 40%+ bench you have seen flying around since thats not the case. It sure will be yet another solid card from the green team, but like i said.. Nothing special.

[kidsafe]

To be fair, the AMD CPUs in general don't necessarily draw more power, per se.
They count TDP-values differently than intel does.
As far as Bulldozer goes, it wasn't that powerconsuming, unless overclocked, which is nothing they expect the general user to do.
The Sandy Bridge-E series also quickly ran away with power consumption, but had a bit more performance and performance gain to back it up, making it a worthwile thing.

They're both being dishonest. AMD claims the Radeon HD 7970 has a max TDP of 250W and a nominal TDP of 210W. At stock volts, my overclocked HD 7970 hits 300W DC. Nvidia claims the GTX 580 has a max TDP of 244W...Hahahah!

As far as FX-Series power consumption is concerned...it's quite a bit worse than Sandy Bridge. All FX chips have 1.2B transistors and a 315mm^2 die size. For comparison, Sandy Bridge has a 216mm^2 die size. You would think that power consumption scales linearly, but it doesn't. GlobalFoundries's 32nm process is apparently terrible and the chips suffer from a ton of gate leakage. As such the FX-8150 consumes even more power than the i7-3960X, which is a 2.27B transistor, 435mm^2 chip...at stock clocks.

[CheezusCrust]

I saw some laptops with Nvidia 630M today, maybe they're getting launched soon-ish?

http://www.komplett.dk/k/ki.aspx?sku=654274

[tetrisGOAT]

I saw some laptops with Nvidia 630M today, maybe they're getting launched soon-ish?

http://www.komplett.dk/k/ki.aspx?sku=654274

http://www.sweclockers.com/nyhet/148...arbara-datorer (Google translate it)
Yea, they've been expected and I can't find the article where actual models been mentioned in the past.
But I believe it's just Fermi-shrinks rather than Kepler.

[Synthaxx]

They're both being dishonest. AMD claims the Radeon HD 7970 has a max TDP of 250W and a nominal TDP of 210W. At stock volts, my overclocked HD 7970 hits 300W DC. Nvidia claims the GTX 580 has a max TDP of 244W...Hahahah!

As far as FX-Series power consumption is concerned...it's quite a bit worse than Sandy Bridge. All FX chips have 1.2B transistors and a 315mm^2 die size. For comparison, Sandy Bridge has a 216mm^2 die size. You would think that power consumption scales linearly, but it doesn't. GlobalFoundries's 32nm process is apparently terrible and the chips suffer from a ton of gate leakage. As such the FX-8150 consumes even more power than the i7-3960X, which is a 2.27B transistor, 435mm^2 chip...at stock clocks.

And yet, chips are built to spec to stay within their power constraints. You disregard TDP and power consumption the moment you go outside those specs by either increasing or reducing the voltage, or by changing the clock speed the chip is running at. And yeah, TDP refers to the amount of cooling potential/energy transfer needed to cool the chip and keep it within it's thermal limits. It makes no exact, perfect correlation or promise between power consumption and TDP, though the values are usually close.