[VR] NVIDIA GeForce GTX 680 Specifications Revealed

[Leadbox]

At this point, we really can't pronounce ourselves on that, though. Early on it made sense that it would be second to GK110, but we have no idea what's up with that. Also, if this is GTX680 and mid range, how is GK110 named, then?

To me, if they're using the GTX 680 name for GK104, it feels a lot like they just pushed back GK110 to the GTX7xx series, which would make sense, especially if GK104 can hold its own vs. AMD's offerings for now! :biggrin:

Nah! My guess is, it only became the GK104 when they realized the performance wasn't quite were the wanted it to be. This bigK or GK110 that's supposedly coming out later is actually a quickfire refresh

 

[GaiaHunter]

The scenarios:

1) The shaders are at least as powerful as Fermi ones and this will be a beast (raises the question if the same bandwidth as GTX580 won't hold them back and how did nvidia improve their xtor density so much ).

2) The shaders are less powerful/might be some kind of vliw like AMD previous designs, so performance is an unknown and could explain the "is faster than a 7970 in some scenarios and slower than a pitcairn in others" rumours (makes more sense than physx).

3) Its a dual GPU and the GK104 is a fermi style 768 shader part, but a dual GPU would be something rumours would pick pe very fast.

 

[f1sherman]

7970 is slower in some scenarios then Pitcairn

my guess shaders are roughly 50% of old ones, and we get close to 2x GTX560 Ti

 

[Crap Daddy]

7970 is slower in some scenarios then Pitcairn

my guess shaders are roughly 50% of old ones, and we get close to 2x GTX560 Ti

That's exactly the point. It's 2x 384 CC GF114 cores (560Ti). That 768 Fermi shaders multiply by 2 and you have the 1536 Kepler cores/shaders.

 

[Vesku]

Perhaps NVIDIA cards will finally be an option for Bitcoin mining, if so I'll be happy to have some diversity in my GPUs.

 

[tviceman]

The conflicting information, only a few days from it's introduction, is making me laugh.

 

[Ketchup]

I guess I'll chime in, why not right? I find it interesting that Nvidia waited for AMD to release new cards, before showing what they have up their sleeve. Could be a strategic move, just to make sure they have something better to show off. They would no doubt have to push back development if they did not, but something tells me they do.

 

[exar333]

In for at least 1 if this is true, and performance is as impressive as the stats.

 

[Red Storm]

The only stat that interests me is the release date. Bring on the price war!

 

[Arachnotronic]

Looks like...it won't be a huge upgrade over a 580. Ah well, FS/FT forum, here I come for some SLI action!

 

[BallaTheFeared]

Some people seem to think this will launch at $650 do to overwhelming performance over the $560 7970....

If that's the case, 28nm, do not want.

 

[Imouto]

Some people seem to think this will launch at $650 do to overwhelming performance over the $560 7970....

If that's the case, 28nm, do not want.

Both AMD and Nvidia will charge a premium for the new gen while having last gen performance.

HD 7970 +20% = GTX 680 +100$
GTX 580 +20% = HD 7970 +100$
GTX 580 = HD 7950 +-0$
GTX 570 HD 6970 = HD 7870 +50$
GTX 560 Ti HD 6950 = HD 7850 +50$

Then remove all your previous gen cards from the shelves and you have a nice trap there. A nice price bump across the entire lineup.

 

[Anarchist420]

If it really has 1536 CUDA cores on one die, then I doubt they could be clocked higher than 1GHz. Say they had 64 ROPs @ 500 MHz, that's less than 1/3 increase in pixel rate over a 580. Then, the zixel rate would only be 11% faster than the 7970 (assuming the 7970 is clocked at 900MHz) This is going to be one big ass chip, so I'm going to wait for a refresh.

One thing I was hoping for was FP64 shader precision in not more than two clock cycles. That would be really good. However, nvidia has severe driver problems, not hardware problems IMO. After the refresh of these, then they should probably shift all of their current hardware resources to the driver team.

 

[Arzachel]

I guess I'll chime in, why not right? I find it interesting that Nvidia waited for AMD to release new cards, before showing what they have up their sleeve. Could be a strategic move, just to make sure they have something better to show off. They would no doubt have to push back development if they did not, but something tells me they do.

It's more likely that they simply had issues transitioning to the 28nm process. Nvidia wouldn't just let AMD have almost the whole market to themselves for a few months if they could've prevented that.

 

[Leadbox]

I guess I'll chime in, why not right? I find it interesting that Nvidia waited for AMD to release new cards, before showing what they have up their sleeve. Could be a strategic move, just to make sure they have something better to show off. They would no doubt have to push back development if they did not, but something tells me they do.

Eh?It is what it is, if its not quite up to it they will sell for less and compete on price, if its faster, u can expect to pay through the nose for it. The only thing they can "develop" now is drivers, anything else would mean several months delay

 

[railven]

Both AMD and Nvidia will charge a premium for the new gen while having last gen performance.

HD 7970 +20% = GTX 680 +100$
GTX 580 +20% = HD 7970 +100$
GTX 580 = HD 7950 +-0$
GTX 570 HD 6970 = HD 7870 +50$
GTX 560 Ti HD 6950 = HD 7850 +50$

Then remove all your previous gen cards from the shelves and you have a nice trap there. A nice price bump across the entire lineup.

That's only relative to those that have last gen parts. For those of us sitting on <= HD5k/GTX 4xx these are impressive gains, with good power usage, lots of headroom, and overall decent price (ie speaking only for AMD side, since of course I can't comment yet on nVidia.)

You guys that just recently bought, kudos - you don't have to buy again

 

[SirPauly]

Both AMD and Nvidia will charge a premium for the new gen while having last gen performance.

HD 7970 +20% = GTX 680 +100$
GTX 580 +20% = HD 7970 +100$
GTX 580 = HD 7950 +-0$
GTX 570 HD 6970 = HD 7870 +50$
GTX 560 Ti HD 6950 = HD 7850 +50$

Then remove all your previous gen cards from the shelves and you have a nice trap there. A nice price bump across the entire lineup.

Let's continue on adding to the 40nm price performance for 28nm product. How will nVidia garner OEM wins with this logic.

 

[thilanliyan]

That things look like it could cost $600-$900.

Nah, Jen Hsun Huang will give us a friend discount.

I really hope the release date is true. Finally, some badly needed competition!

 

[BallaTheFeared]

Nah, Jen Hsun Huang will give us a friend discount.

I really hope the release date is true. Finally, some badly needed competition!

I think AMD might have set the market in so far as Nvidia is going to price based on AMD's already established pricing... Assuming they have card(s) faster than teh 7970 and they feel they can sell it at $600+ that's where they'll price it.

 

[blckgrffn]

I'll believe it when I see it. It's the safest way to roll

 

[Covrich]

Whilst the Stats look good what I am worried about is Nvidia will base pricing over any and above AMDs 7xxx Series which has of course taken straight over from % over last generation.

We could see a $650 Card here and if prices do not drop where will it end.

I am in the market for a new card since my old one went But I am reluctant to part with the cash yet until we know what is what.

At the minute the 7870 is looking tempting but I am not going to under estimate Nvidia. we need some pricing competition now!

 

[tviceman]

Here's a big fat juicy interview (in Japanese).

http://www.4gamer.net/games/120/G012093/20120306077/

I think the details are want to tell once again the opportunity, as can be seen if Moraere Think of a Hyper-Threading "" of Intel "approach is NVIDIA, to be able to use threading to other free resources CUDA Core It is, and shall raise the processing performance per CUDA Core, be reduced so that the power consumption per performance "that (official NVIDIA).

Unfortunately since the original text/interview is in Japanese, it's hard to tell exactly what the interview is saying. So definitely take this with a big grain of salt, but it could explain quite a few things - like the contradicting specifications rumors and how Nvidia might have managed to get such a huge performance bump on a chip smaller than GF104 and Tahiti. As we all know, GPU's are highly threaded processors so creating some sort of hyperthreading environment might have way more performance benefits on GPU's than on CPU's.

 

[RussianSensation]

I think AMD might have set the market in so far as Nvidia is going to price based on AMD's already established pricing... Assuming they have card(s) faster than teh 7970 and they feel they can sell it at $600+ that's where they'll price it.

To me this smells like 7800 GTX launch all over again.

The "fake" flagship - 7800GTX - Launched June 2005
24 pixel, 8 vertex, 24 TMUs
430mhz GPU
256MB, 1200 DDR memory (38.4GB/sec bandwidth)

Then 5 months later

The "Real" flagship successor to 6800 Ultra - 7800GTX 512MB - Launched November 2005

550mhz GPU (+28%)
512MB, 1700 DDR memory (54.4GB/sec bandwidth = +42%)

Why do I get this feeling that GK104 will be a $550+ GPU and NV will launch the "real" Kepler flagship with 384-bit, 2304 SPs as the GTX780 6-7 months later...?

If this 320-340mm^2 256-bit chip is sufficient enough to beat GTX580 by 50%, then the Big K could easily add another 40-50% on top of that and NV might as well call that the next generation GTX780

 

[TestKing123]

The 7800GTX 512 was vaporware. It was a response to the X1800 series and was produced in so limited quantities that it was an industry joke.

 

[nitromullet]

That's true Russian, but the availability of the 7800GTX 512 was so scarce that most of us with 7800GTXes waited until 7900GTX to upgrade. I know I did.

GK114 will probably be faster then my current card, which will be eligible for an EVGA step-up all the way through April, so I'm most likely in for one. If they sell for $350-450, maybe two.