Kepler Nvidia GeForce GTX 780

[Lanek]

No official info tells you exactly that, no official info, nothing more or nothing less.

If NV released some sort of sneak peak, you would discount the results as cherry picked. If they paper launched, rumors of magical driver or higher clocked BIOS update would surface in a matter of minutes...

By now, both companies probably learned, there is just no pleasing some of the people.

I dont speak about performance, but as what is the architecture... this is why i was made a comparaison with AMD who have show us GCN architecture in June 2011... Im sure a lot of peoples in server, computing, professional graphism will be interested to know what this new architecture can bring as feature on the table.. ( starting by me for my job... ( well im a DAO, CAO designer Autocad )

[Olivon]

Lol, does OBR remain serious ? this guy should have 2 bodies for wear a so big head.

How can he have any idea of the Kepler performance ( he speak about it since 2 month now ), if the A1 silicon was bugged, and the A2 sillicon have been out of TSMC less of 4 days, 1 week ago?

A1 was bugged ? really ?

[Lanek]

A1 was bugged ? really ?

Yes, but you offtly see an A1 who is not bugged for nvidia ? ( whatever is the rumors about it, they dont take any risk, ( interconnect who was not working well ).

[Tha Last Meal]

Lol, does OBR remain serious ? this guy should have 2 bodies for wear a so big head.

How can he have any idea of the Kepler performance ( he speak about it since 2 month now ), if the A1 silicon was bugged, and the A2 sillicon have been out of TSMC less of 4 days, 1 week ago?

lol man ......

Why don't you forward these questions to Charlie or everything said by Charlie is true and by others are not, you have to build your analysis on data and basic information not on personal point of view..if I take your word then that means what others say other than Charlie is false and not logical which is absurd..be more reasonable and logical.

[BHD]

Nvidia Kepler GTX690

750MHz Core clock
2×1.75 GB 4.5GHz GDDR5 Memory
2×1024 Stream Processors
2x448bit Bus Width
Priced at $999

-----------------------


Nvidia Kepler GTX680

850MHz Core clock
2 GB 5.5GHz GDDR5 Memory
1024 Stream Processors
512bit Bus Width
Priced at $649

------------------------


Nvidia Kepler GTX670

850MHz Core clock
1.75 GB 5GHz GDDR5 Memory
896 Stream Processors
448bit Bus Width
Priced at $499


-------------------------


Nvidia Kepler GTX660Ti

850MHz Core clock
1.5 GB 5GHz GDDR5 Memory
768 Stream Processors
384bit Bus Width
Priced at $399


--------------------------


GTX 680 45% Faster (Radeon HD 7970)

[japamd]

Nvidia Kepler GTX690

750MHz Core clock
2×1.75 GB 4.5GHz GDDR5 Memory
2×1024 Stream Processors
2x448bit Bus Width
Priced at $999

-----------------------


Nvidia Kepler GTX680

850MHz Core clock
2 GB 5.5GHz GDDR5 Memory
1024 Stream Processors
512bit Bus Width
Priced at $649

------------------------


Nvidia Kepler GTX670

850MHz Core clock
1.75 GB 5GHz GDDR5 Memory
896 Stream Processors
448bit Bus Width
Priced at $499


-------------------------


Nvidia Kepler GTX660Ti

850MHz Core clock
1.5 GB 5GHz GDDR5 Memory
768 Stream Processors
384bit Bus Width
Priced at $399


--------------------------


GTX 680 45% Faster (Radeon HD 7970)

Wow, 45% faster is nice, but priced at 649?

And OFC, where's the 299 Kepler that would crush Tahiti? :P

[cold2010]

this ?

http://lenzfire.com/2012/02/entire-n...se-date-43823/

[informal]

@BHD
And your source is? Not that I don't believe you or anything,but 7970 is ~34% faster than 580GTX in 25x16 with 4xAA/16xAF. In order to be 45% faster than 7970,single GPU 680 needs to perform around 1.93x better than 580GTX which even though is not impossible ,is highly unlikely. You have 2x the "stream" processor count but you have roughly the ~2x less shader clock which is now in sync with the rest of the GPU. You have almost 84% more memory BW though so there is close to enough of it for this to be true. But SPs have to be massively more efficient to net such a huge gain over GF110.

[BHD]

My Friend Informal

Source >
http://lenzfire.com/2012/02/entire-n...se-date-43823/

[Andi64]

Wow, 45% faster is nice, but priced at 649?

And OFC, where's the 299 Kepler that would crush Tahiti? :P

That would be the GTX660Ti, and it looks it will be priced U$S399. Based on what BHD said, if a GTX680 is 45% faster than HD7970, that would mean a GTX660 will be ~5-10% faster.

The price looks pretty nice if you ask me. And it also makes sense. It has 33% more SPs than a GTX580, if you factor in architecture optimizations... it should be faster than HD7970 (which is 25% faster than GTX580 on average)

That's why a new AMD sku makes sense. Every HD7970 is clocking 1200Mhz easily, thats 30% more clock, that usually translates to 25% more performance. That hypothetical "HD7980" would perform between GTX670 and GTX680. Just what AMD has been doing all these last years...

[/Speculation]

[nossy23]

Looks really good. I hope it's true. Still, only 2 GB memory

[Manicdan]

a chip with 1024 SPs would be probably 20% bigger than a 580, and that took a year to come out and why we had the castrated version called a 480

[ajaidev]

Interesting according to the website GTX660Ti and up all are 550mm2 so that means its the same chip, 7970 is about 365 mm2. GTX660Ti does seem to be the logical fighter for 7950 not 7970 that 10% better than 7950 might be on specific tests. I do expect GTX670 to exceed 7970's performance and GTX680 to destroy 7970 "If the charts on the website are true"

[mAJORD]

@BHD
And your source is? Not that I don't believe you or anything,but 7970 is ~34% faster than 580GTX in 25x16 with 4xAA/16xAF. In order to be 45% faster than 7970,single GPU 680 needs to perform around 1.93x better than 580GTX which even though is not impossible ,is highly unlikely. You have 2x the "stream" processor count but you have roughly the ~2x less shader clock which is now in sync with the rest of the GPU. You have almost 84% more memory BW though so there is close to enough of it for this to be true. But SPs have to be massively more efficient to net such a huge gain over GF110.

In theory, I don't think it's that unreasonable for it to approach double Fermi performance (Say +80% ) with those spec's (and accompanying huge die size). What I do think would be a problem in reality is thermal headroom. whilst you can double up your shader count from one node to another, it usually accompanies a marked power consumption increase, and even fermi 2 (GTX580) is pushing the boundries of what's acceptable (With the GTX480 clearly demonstrating what's not acceptable)

So IMO what would make or break it is how much they've improved their performance/watt of the architecturen, as they've been behind AMD for some time in this regard (though not by much )

I think 60-70% faster would be more realistic expectation, which would still make it considerably faster than the 7970. I don't think AMD would respond with anything wild though of course. They have no interest in pushing limits in die size and power on GPU's.. Their strategy seems to be release something early, and enjoy a comfortable lead for ~6mths, then sit in the sweet spot when Nv finally releases their monster.

A modest successor / refresh to the conservativley clocked 7970 would be enough to maintain them in their previous position for single GPU I think (6970 vs 580, (5870 vs 480) (4870 vs 280) etc

[tajoh111]

Really hard to believe. Too much information about the entire line up too soon.

We have never known really that much information about a a lineup this early. Especially performance numbers and most importantly, accurate numbers.

[jimbo75]

Not to mention Nvidia suddenly being ultra-coy about their amazing performance

[GoldenTiger]

I'm impressed if this stuff's true. Still, two months in advance of launch, seems... suspect?

[jimbo75]

Come on really, if they had this kind of performance they'd show it. Just one official benchmark - they did it with Fermi 4 months before the 480 was released, showing how it pwnd the 5870 in Far Cry 2. Surely they can show how amazing they are in any title right now? Just 1 official benchmark, that's all we ask.

[Lanek]

Interesting according to the website GTX660Ti and up all are 550mm2 so that means its the same chip, 7970 is about 365 mm2. GTX660Ti does seem to be the logical fighter for 7950 not 7970 that 10% better than 7950 might be on specific tests. I do expect GTX670 to exceed 7970's performance and GTX680 to destroy 7970 "If the charts on the website are true"

Whatever if this is true or not, the TI version is what the TI version was for the 560TI, a 470 or if you like more a castrated 570-580.. I doubt they want to launch something like that, the 560TI was a card for use bad cores who cant be used on 570-580. other than that, this is a serious waste of money, take a 550mm2 chips cost and laser cut a good part of it..

You do that, when you are sure to have xxxxx numbers of core who cant pass the test for been on the 670-680 ... you dont think do it before that. Cause you have no idea of how many cards you will product. If you need take the good cores who can be send on the 670-680 line, you loose a lot of money and wafers.

[Handrox]

GTX 680 45% Faster (Radeon HD 7970) But 1000$ price for DUAL GPU

[NBF]

GTX 680 45% Faster (Radeon HD 7970) But 1000$ price for DUAL GPU

Seems too good to be true!
I think the only real infos are the prices

[Handrox]

Seems too good to be true!
I think the only real infos are the prices

1K is the suggested retail price, imagine final price ....

[Levish]

Lordy, gonna need a 2nd job for my next gpu upgrade.

[Chrono Detector]

If those specs are true, it looks nice, but again, whats with the low VRAM? Seems like NVIDIA did not learn their lesson from GTX 580.

[iboomalot]

64 ROPs thats alot

7970 is only 32 ROPs I think AMD needs to up those and not wait 4years like they did going up from 16 rops in the past