[ExtremeTech] GTX 680 "...will probably cost upwards of $500."

[Quantos]

Or we could say everything screams high end:
The price
The launch timings (usually high end cards launch first in a new series. Not always, but usually).
The RAM speed and amounts. 4GB on a mid-range card? What?
The name - GTX680 is the high end card name.

This.

This card really doesn't feel midrange, or low high end. A lot of things about it just scream high end. The name itself is enough to convince.

 

[BallaTheFeared]

Or we could say everything screams high end:
The price
The launch timings (usually high end cards launch first in a new series. Not always, but usually).
The RAM speed and amounts. 4GB on a mid-range card? What?
The name - GTX680 is the high end card name.

Sadly.

We don't have the full list of cards, they could also be playing it safer this time around since they've been paying per wafer... A smaller die chip like a GF104 460 would produce many more chips per wafer than GF100. Clearly they would want to avoid another Fermi deal, lower yields with a smaller die still means more total yield than if they were using a larger die.

192GB/s is the same as the GTX 580, no headway was made. Same speed on 512 bit bus would be considerably more impressive. 4GB comes later reportedly, like a 2GB 460 came later, like a 2GB 560ti came later, like a 2.6GB 570 came later, like a 3GB 580 came later, doesn't mean anything.

We'll see on the name, could be the card we're talking about, or that could be the 670ti which is also rumored, dunno yet.

Quit speaking to yourself? There are others here to talk with.

I speak for the trees, if I don't speak for them who will?

Considering NVIDIA apparently will release a dual GK104 and according to the rumours, you might have to wait.

I had already planned to wait, go premiums... I was just hoping the bigger card would come out sooner so the refresh would happen sooner as well.

 

[PingviN]

Why should they do it? They have no problems with huge Dies.

Fermi was 6 months late, barely better than Cypress that's gotta count as a problem.

 

[sontin]

Fermi was 6 months late, barely better than Cypress that's gotta count as a problem.

Because of a design flaw and not of the die size.

 

[Arzachel]

Because of a design flaw and not of the die size.

I think I've disagreed with you on this before, but still,what is this design flaw you speak of? When your die is so large that you can't get any chips to yield without fusing off execution units, the size of the die is very much a "design flaw".

 

[BallaTheFeared]

lol...

They had the a bigger die on 65nm

 

[3DVagabond]

lol...

They had the a bigger die on 65nm

What does that have to do with 40nm, or now, 28nm? 40nm waslate and required a refresh before they could get their designs working 100%. They are having to release a smaller die first on 28nm and it's projected many months down the road before their big die comes along. Connect the dots.

 

[Arzachel]

lol...

They had the a bigger die on 65nm

Which they made on a very mature node they had been using for 7+ months. Your point?

 

[BallaTheFeared]

Context doesn't hurt...

My point was big die isn't a design flaw, it's kept them on top for awhile now.


Too much smack talk about die size from the AMD crowd though, we'll see what happens now that AMD stepped up to Nvidia's level with tessellation performance and GPGPU.

Rumor has it the GK104 chip is smaller, faster, and uses less power... We'll see though.

 

[Arzachel]

Context doesn't hurt...

My point was big die isn't a design flaw, it's kept them on top for awhile now.

Did you even read my post?

When your die is so large that you can't get any chips to yield without fusing off execution units, the size of the die is very much a "design flaw".

While it might kept them on top before, it's because they've had the luxury of not jumping on every new node as soon as possible, because AMD/Ati was so far behind.

 

[GaiaHunter]

I don't think big die is a bad decision per se.

It depends on the circumstances.

There is a reason that for while we would read on these forums "new architecture on a new process is recipe for disaster".

But these days new process seems to take longer to arrive and seems more difficult to get them to maturity.

So companies need to adapt.

Small dies isn't better per se either.

AMD either was lucky, had a better forecast, or something in between and in the end their design philosophy allowed the to close the gap and go back and forwards trading places with NVIDIA for top spot.

 

[Arzachel]

I expect AMD to alternate between "small" die on a new process and a larger die once the process matures. Looks like Nvidia will be doing something like that but that's only because they're forced to.

 

[Ieat]

If its a penny over $450 nivida can shove it. Might as well get a 7950 with the extra vram or a 7870 for a bill less. Not that I would get any of these cards as the value ranges from average to piss poor.

 

[BallaTheFeared]

Did you even read my post?



While it might kept them on top before, it's because they've had the luxury of not jumping on every new node as soon as possible, because AMD/Ati was so far behind.

I was under the impression that Nvidia had a lot of problems with leaky transistors going full size would have pushed the power draw too far and it already took major "heat" for that on release.

The problem wasn't the big die, the problem was the design itself.

They actually never really solved the design problem, or so I was lead to believe. They actually just put insulators per say around the leaky problematic transistors and cut them off when they released the GTX 580, which is why the 580 has 200,000 less transistors and an slightly smaller die size.


Is that wrong, was the problem the die size, or was it the design?

 

[BlockheadBrown]

If true, it just gives many people more reason to wait. The card won't be as desirable for those who were hoping for a lower price point. ::shrug:: Ultimately, that price helps Nvidia more than it hurts. For me personally, I'm looking at the sub-$200 range. It'll be a while before I upgrade. Even then, it'll be a cautious move. AMD may have their refresh out by October/November. I'm (totally) guessing Nvidia wouldn't have a decent refresh for the sub-$200 range until roughly Feb of 2013. ::shrug:: This may give me more reason to look/hope for a sub-$150 560 over the next few months. April should be pretty telling for what I'm looking for.

 

[blackened23]

Chinese folks already have the card, has been benchmarked:
http://forum.beyond3d.com/showpost.php?p=1626092&postcount=2212

Translated

already got the GK104 kind of card , the stream processors 1536, about 5% lower than the HD7970 in performance, seconds away relatively satisfied with the HD7950, temperature and noise, close to GTX560ti It is said that the price 3299-3499


35000 yen is 430$ US dollars

 

[BallaTheFeared]

Says GTX670Ti, which was reportedly between the 7970 and the 7950, what about the 680?

 

[SlowSpyder]

B.S. opinion article. Everyone knows the GK104 is a $299 7970 killer.

 

[blackened23]

Says GTX670Ti, which was reportedly between the 7970 and the 7950, what about the 680?

The lower GK104 doesn't have 1536 stream processors. The part we thought was the 670ti turns out to be the 680.

Note: I do not know if the figures are true, i'm guessing that the GK104 will have peaky performance.

 

[Lonyo]

Chinese folks already have the card, has been benchmarked:
http://forum.beyond3d.com/showpost.php?p=1626092&postcount=2212

Translated

already got the GK104 kind of card , the stream processors 1536, about 5% lower than the HD7970 in performance, seconds away relatively satisfied with the HD7950, temperature and noise, close to GTX560ti It is said that the price 3299-3499


35000 yen is 430$ US dollars

Since it's a Chinese leak, the 3300-3500 probably refers to Chinese yuan.
3300RMB = $522 USD.
Which doesn't mean it will be $522 USD, since prices don't translate like that, but added to the original $500 mention suggests it could indeed be a $500 card.
5% slower for $50 less. Price/$ would be better, but it won't put any real pressure on any of the market if true.

 

[SlowSpyder]

I cannot believe anyone is trying to argue this is a midrange part if it is getting a 670/680 label and selling for $500 (going on rumors here). There may be a higher end part coming, but it is very likely not coming very soon.

edit - Just wanted to add, GO MARGINS!

Man, if this thing is $500+ and slower than the 7970 or even marginally faster, there are going to be so many people eating crow...

 

[BallaTheFeared]

I cannot believe anyone is trying to argue this is a midrange part if it is getting a 670/680 label and selling for $500 (going on rumors here). There may be a higher end part coming, but it is very likely not coming very soon.

Believe it because it's happening \>Right Now</

 

[GaiaHunter]

I cannot believe anyone is trying to argue this is a midrange part if it is getting a 670/680 label and selling for $500 (going on rumors here). There may be a higher end part coming, but it is very likely not coming very soon.

Well (some) people were expecting the GTX680 to be twice as fast as the GTX580.

The price and price/performance of this GTX680 GK104 will be a cue on when the GK100/110 will be released - if NVIDIA try to significantly undercut AMD it will mean a faster card will arrive sooner than later, otherwise those Q3 and later projections will be more probable.

 

[railven]

Quit it with the timeframe-shifting "midrange" nonsense. If you're going to claim that the high end Kepler coming out several months from now is the "real" high end Kepler and the GTX 680 is "midrange," fine, but by that token the HD 7970 is also a "mid range" part and its refresh (HD 8970 or HD 7980 or whatever) is the "high end."

Midrange, high end, etc. are meaningless when you shift timeframes by more than a month or two.

Dollars and cents and product position in the near term are what matter.

$500+ = GTX 680, HD 7970
$450-500 = GTX 580, HD 7950
$350-400 = GTX 570, HD 7870, HD 6970 2GB
Anything less = might as well get a GTX 560 Ti or HD 6950 instead and save some cash for a future upgrade

Lackluster choices all around. :/

Well said guy

On the low end, it seems those cards are going to start disappearing soon. I'd personally wait a tad longer for the channel to dry up and hopefully the current cards (HD 78xx/GTX 6xx whatever) will start to slot into those price points.

I'd personally advice against buying last gen technology even more so when the price difference are so marginal (again, this doesn't include clearance prices, I mean more so the usual going price.) But that's just me.

 

[boxleitnerb]

I fear it will be indeed Q3 or Q4. Nvidia might decide to milk the cow, too. If GK104 is sufficient to battle Tahiti XT, they might sell all their first GK110 chips to the HPC clusters and as Quadros and then when the yields have improved considerably or when they run out of customers for the time being, they bring it as GTX780.