[3dcenter] GK104 specs

[Stuka87]

This is how I am seeing it as well. They hit the same problems with getting a massive die on a new node out in a timely fashion and have pushed it back, but rather than wait till it's ready, they'll release a smaller and more easily made chip sooner; GK104.

My belief is that they'll release the card as a 680 that can compete with a 7970 with a die size similar to Tahiti, charge at least $450-$500 for it, and put out a few cut down models for $250-$400.

Late this year they'll release the huge GK110 chip to compete against an AMD refresh of Tahiti, at which point they'll solidly take the performance crown, unless AMD breaks the mold and decides to make a huge die as well, which I don't see happening.

The one steaming turd is that the full-fledged GK104 is going to cost $300, this is not going to happen, but I do see a cut down model at that price point. This is nvidia, they put out a card that is X80 model line, you will be paying a fat price for it.

This totally makes sense to me. And if the GK110 is as far out as rumors say, it would make sense.

 

[notty22]

NVIDIA to Demonstrate PhysX on GK104, Kepler GPU on Friday?

The very latest information comes from EVGA, Gearbox Software and NVIDIA. This weekend will see 19th PDXLan taking place in Portland, Oregon. Over the course of weekend, EVGA plans to showing the next generation "Kepler" running PhysX demo of a new game from Gearbox Software.
We expect that specifications stay the secret over the course of weekend, with everybody talking about the game alone, and the official unveil of Kepler will follow in weeks to come.

​

 

[Lonyo]

Yes, and how did that end up? A $150 price drop 2 weeks after release, $125 and $60 rebates handed out to GTX 280/260 owners. This time 7970 buyers will end up with a $150 hole in their pocket.

The whole point of this thread is that no one knows for sure how it will perform or what the price will be... it could be a $150 hole, it could be less (if the GTX6x0 performs slower than an HD7970 but is priced cheaper) or it could be more.

And 2 weeks is a little different to 2 months (and probably more).

 

[PrincessFrosty]

Honestly the impressiveness of the mid range chip isn't really important to me, if they were launching both the mid range and the top tier together then relatively the top end is going to be a real impressive piece of hardware but if it's going to be late then who really cares, by then we'll be comparing it to an AMD refresh.

 

[Keysplayr]

NVIDIA to Demonstrate PhysX on GK104, Kepler GPU on Friday?

And what they said here was most interesting as well:

"First and foremost, Samsung showed an Ivy Bridge + mainstream Kepler GPU notebook on CES 2012 which got its official page shown by accident"

So that wasn't a mistake. Many thought it was a Fermi based mobile chip in that lappy.

Kepler is closer than we think IMHO.

 

[boxleitnerb]

GK104 will come in March, GK110 in May/June. My personal guess.
Isn't it time to make a poll regarding performance and launch date?

 

[blackened23]

GK104 will come in March, GK110 in May/June. My personal guess.
Isn't it time to make a poll regarding performance and launch date?

Okay, GK104 definitely would not be the 680 if GK110 were anything close to a june release. While it would be nice, I would definitely buy it based on the specs alone - GK110 isn't in A1 status yet nor has it been validated. I'm thinking what everyone else is stating - Q3 _if_ it doesn't have problems along the way. (has nvidia ever had a problem free part??)

 

[boxleitnerb]

With Nvidia, there is no A0. A1 is the first stepping.
Names don't have to represent actual performance, see 6870 vs 5870. 680 could be named just that for marketing reasons because this time around the big chip was always supposed to come after the smaller one.

Q3...like October? 10-11 months from tape out to launch, really? It took 6 months (including 2 respins) for Fermi. Please stop kidding

 

[blackened23]

With Nvidia, there is no A0. A1 is the first stepping.
Names don't have to represent actual performance, see 6870 vs 5870. 680 could be named just that for marketing reasons because this time around the big chip was always supposed to come after the smaller one.

Thats a good point, nvidia hasn't always followed their naming scheme to a tee (neither has AMD for that matter). I'm thinking that nvidia will clock the 680 as high as humanly possible to put it in the best possible light as to compare favorably to the tahiti series....

Man. I really want to make a road trip to the PDX lan thing this friday and catch a glimpse. It doesn't help that I'm a huge borderlands fan -- I presume they're showing borderlands 2 off. This is making me want to give in to the dark side of the force

 

[boxleitnerb]

Maybe there will be some videos? After all, this event is kind of public, right?

 

[Arzachel]

With Nvidia, there is no A0. A1 is the first stepping.
Names don't have to represent actual performance, see 6870 vs 5870. 680 could be named just that for marketing reasons because this time around the big chip was always supposed to come after the smaller one.

Q3...like October? 10-11 months from tape out to launch, really? It took 6 months (including 2 respins) for Fermi. Please stop kidding

GF100 taped out June 16(BSN) or July 29(Charlie) depending who you trust. You've been corrected about that 6 month figure once before in this very thread.

 

[tviceman]

Sounds like you're hedging your bets already tviceman without knowing anything concrete

It sounds like you don't want to take that friendly bet! You're right, I don't know anything more than anyone else here, but I do know this: Fermi's protracted development and rumors were generally negative while GK104's rumors have been very,very positive. We each draw our own conclusions.

 

[boxleitnerb]

GF100 taped out June 16(BSN) or July 29(Charlie) depending who you trust. You've been corrected about that 6 month figure once before in this very thread.

Corrected by what authority? And If I trust neither one? I was wrong about December, I give you that. In the meantime I have been asking around because I want to know what is realistic for GK110:
Ailuros from 3DC who has a better track record and he said GF100 tape out was late August 2009. I rather believe him than BSN or Char_LIE.

The time from tape out to launch/availability usually is 3-6 months, depending on the complexity of the chip. 10 months? Give me a break...

 

[Arzachel]

Corrected by what authority? And If I trust neither one? I was wrong about December, I give you that. In the meantime I have been asking around because I want to know what is realistic for GK110:
Ailuros from 3DC who has a better track record and he said GF100 tape out was late August 2009. I rather believe him than BSN or Char_LIE.

The time from tape out to launch/availability usually is 3-6 months, depending on the complexity of the chip. 10 months? Give me a break...

...3 months? Well you could ask IDC what's the ETA on base and metal layer respins . Even rv770 which iirc was released on A0 silicon took ~6 months.

 

[Lonyo]

Corrected by what authority? And If I trust neither one? I was wrong about December, I give you that. In the meantime I have been asking around because I want to know what is realistic for GK110:
Ailuros from 3DC who has a better track record and he said GF100 tape out was late August 2009. I rather believe him than BSN or Char_LIE.

The time from tape out to launch/availability usually is 3-6 months, depending on the complexity of the chip. 10 months? Give me a break...

3-6 months?
So when AMD demonstrated cards in June that they wouldn't launch until January, that's 6 months from working silicon to production.
And according to sites it had taped out in or before April.
That's over 8 months.

 

[boxleitnerb]

With more complex chips and processes it obviously is tilted towards the 6 months.

Take G80 for example. Tape out was around June 2006, launch was beginning of November. 6 months is possible imo. Some people here are talking about Q3 or even Q4 launch. Wouldn't you agree that this is just ridiculous?

So when AMD demonstrated cards in June that they wouldn't launch until January, that's 6 months from working silicon to production.
And according to sites it had taped out in or before April.
That's over 8 months.

You're forgetting something here:
When did mass production start for 28nm in actual volume so you can launch a product in the thousands, not dozens? You can tape out early on a new process, but you have to wait until capacity is there to make your chip. AMD and Nvidia are not the only 28nm customers you know? Given the high voltages/low frequencies of Tahiti, 28nm yields aren't that good. But right now, 28nm is running and improving, volume is increasing. So Nvidia doesn't have to wait. If all works well, 6 month are reasonable.

 

[Tempered81]

Regarding Nvidia Geforce Kepler release dates, SemiAccurate wrote in a September 2011 article that the first Kepler part would release near April 1 2012, and that the big Kepler GK100 would have to likely need another resspin or two and put it's availability to mid-2012.

old article here:
http://semiaccurate.com/2011/09/20/exclusive-nvidia-has-two-keplers-in-house-but-not-the-big-one/

 

[Arzachel]

With more complex chips and processes it obviously is tilted towards the 6 months.

Take G80 for example. Tape out was around June 2006, launch was beginning of November. 6 months is possible imo. Some people here are talking about Q3 or even Q4 launch. Wouldn't you agree that this is just ridiculous?



You're forgetting something here:
When did mass production start for 28nm in actual volume so you can launch a product in the thousands, not dozens? You can tape out early on a new process, but you have to wait until capacity is there to make your chip. AMD and Nvidia are not the only 28nm customers you know? Given the high voltages/low frequencies of Tahiti, 28nm yields aren't that good. But right now, 28nm is running and improving, volume is increasing. So Nvidia doesn't have to wait. If all works well, 6 month are reasonable.

Fine, provide an example of a reasonably large(>200mm2) GPU going from tapeout to availability in under 4 months. You do know that even after the chip is done, supplying retailers takes atleast a month, right?

 

[boxleitnerb]

All I'm saying is that 6 months is reasonable if there are no major problems. If some of you believe GK110 will come in Q3 or Q4, we'll see. I won't split hairs with you.

But if you insist, G70 (300+ mm2):
Tape out reported late April (Inquirer link not working anymore)
http://www.xtremesystems.org/forums/showthread.php?59976-INQ-NVIDIA-tapes-out-G70-silicon
It's not entirely clear when exactly the tape out occured, but I bet the 4 months isn't too far off.

Launch on June 22nd.

So, and now btt please.

 

[blackened23]

That isn't quite clear enough for me. How about this. Based on the gaming (nonsynthetic) benchmarks Anandtech runs, I say the top end GK104 card (whatever it is called) will perform on average within 5% of an hd7970 and will have an MSRP of $349 or lower. If the MSRP is a penny higher or the hd7970 is on average at least 5.001% faster than Anandtech's benchmarks then you win.

$30 steam game of winners choice.

How does this sound?

Hmm, why not I'm in I guess.

 

[SirPauly]

Really?
Is it really? Really really?


GTX280 says "Hi!".

First result on Google for "GTX280 MSRP" http://arstechnica.com/hardware/reviews/2008/06/nvidia-geforce-gx2-review.ars/8

(The 9800 was simply a die shrunk 8800, so while you could argue that the GTX280 was the same node as the 9800, it was the first new architecture on that new node, and was competing against an older node designed card. NV also chose to use an older process rather than transitioning to the newest one for their high end GPU, partly due to timings, probably, and also due to yield concerns, so in fact it could have been a 55nm GPU on launch vs a GPU designed at 90nm).

Hehe, there ya go -- let's not forget the GeForce 8800 Ultra as well. So, I guess AMD is now included in the extreme premium race with nVidia now -- congratulations -- but oddly, don't recall anyone defending strongly those nvidia extreme premiums, except maybe Rollo -- more like insane asking so much.

 

[KCfromNC]

Yes, and how did that end up? A $150 price drop 2 weeks after release, $125 and $60 rebates handed out to GTX 280/260 owners. This time 7970 buyers will end up with a $150 hole in their pocket.

Well, at least we know it will take more than 2 weeks in this case. There's a big difference between a $150 price drop if the competition arrives 2 weeks after release compared 6 months after release. We'll see which one is closer to the reality just as soon as we get any concrete information about Kepler.

 

[Arzachel]

All I'm saying is that 6 months is reasonable if there are no major problems. If some of you believe GK110 will come in Q3 or Q4, we'll see. I won't split hairs with you.

But if you insist, G70 (300+ mm2):
Tape out reported late April (Inquirer link not working anymore)
http://www.xtremesystems.org/forums/showthread.php?59976-INQ-NVIDIA-tapes-out-G70-silicon
It's not entirely clear when exactly the tape out occured, but I bet the 4 months isn't too far off.

Launch on June 22nd.

So, and now btt please.

I think he's talking about "tape out of final production silicon", not "initial tape out after RTL completetion" (which I what I think most of us take tapeout to mean). I believe the code on the PCI SIG list is for G70 and it make sense to validate an engineering sample if it is not going to require any changes from the production silicon.

Hehe, there ya go -- let's not forget the GeForce 8800 Ultra as well. So, I guess AMD is now included in the extreme premium race with nVidia now -- congratulations -- but oddly, don't recall anyone defending strongly those nvidia extreme premiums, except maybe Rollo -- more like insane asking so much.

The difference is that AMD went from relatively cheap to expensive. The 8800 Ultra and the 280GTX went from expensive to "I don't have that many kidneys to sell"

Anyways, very best way this could play out, is that Nvidia releases a GTX670 that is just a lower clocked GTX680, without any disabled shaders. I'd get something like that if the price would be right.

 

[SirPauly]

Imho,

50 percent raise in MSRP and raising the MSRP percentage more than the performance percentage is extreme as well. Oddly, it may take nVidia to bring prices down from this extremism because without competition, nVidia and AMD are predators and aggressors.

 

[OILFIELDTRASH]

The difference is that AMD went from relatively cheap to expensive. The 8800 Ultra and the 280GTX went from expensive to "I don't have that many kidneys to sell"

Anyways, very best way this could play out, is that Nvidia releases a GTX670 that is just a lower clocked GTX680, without any disabled shaders. I'd get something like that if the price would be right.

Not sure about you but any price increases suck to me. Whether it be nvidia, amd, intel.. Just from the little I understand about gpu/CPU manufacturing it appears that the 7970 is a relatively cheap gpu to make. So that means that the price is artificially high for no reason but greed. If there was a reason for it to be so high such as manufacturing or whatever then fine. You just suck it up and pay for it. It's things like this that turn me completely off of a company/product. Doesn't help any knowing that the amd CEO sees his customers as prey.