Kepler Nvidia GeForce GTX 780

[AliG]

I'm liking mine more
nVidia-GeForce-GTX-680-Specifications.gif

[bamtan2]

thats how accurate this entire thread has been

[phantomferrari]

only 8.26" long, pff. i need a card atleast 10" before i can get any enjoyment out of it.

i know this is extremely childish and immature but i simply cant resist saying this. Thats what she said!!

im sorry, i will not revert back to my normal age

[Andrea deluxe]

GTX680 = GK104
3*32SPs/SM
8 TMUs/SM
4 SMs/GPC
4 GPCs
32 ROPs
256bit bus
2GB 5.0Gbps GDDR5
950MHz core
2*6pin
________

http://forum.beyond3d.com/showpost.p...postcount=1538

[thatdude90210]

Someone found the originals from that OBR fake. Basically background image of a GTX550 and the benchmark slide from a GTX580.

http://forum.beyond3d.com/showpost.p...postcount=1546

[AliG]

i know this is extremely childish and immature but i simply cant resist saying this. Thats what she said!!

im sorry, i will not revert back to my normal age

I thought that was the point of the joke lol

[Cold Fussion]

If that is true, I'm disappointed it isn't coming with a displayport connection.

[BababooeyHTJ]

Who cares about the wording. It is late. Late to the market, late to beat the competition. It's been 15 months since GTX 580 release, and it's going to be 20 months in Q3.

Also, Nvidia isn't known for great price-performance. While GK104 may compete with 7950, it will be quite expensive if so, with GTX 580 also staying expensive (their high-end...), 7950/70 staying expensive, and 7990 being insanely expensive. And this may last till Q3 thanks to Nvidia...

The majority of their marketshare doesn't care about $500 cards. AMD hasn't effected the sweet spot price on any current gen card. They released a nice enthusiast card for multi-monitor users but this isn't like last gen. Nvidia is competing just fine with ever card in there lineup but GTX580.

[Chrono Detector]

So NVIDIA is going to name the mid range GTX 680, fail naming scheme, reminds me when they used the 9800 GTX name as a mid range card.

[Lanek]

So NVIDIA is going to name the mid range GTX 680, fail naming scheme, reminds me when they used the 9800 GTX name as a mid range card.

you should read the post 555 here.

Someone found the originals from that OBR fake. Basically background image of a GTX550 and the benchmark slide from a GTX580.

http://forum.beyond3d.com/showpost.p...postcount=1546

if you want i can make it GTX660 or any name if you like. (Thanks to ManofAtlantis on beyond3d)


and the slide for the chart is exactly one used when 580 have been released ... I had suspect something yesterday and so i have just find this one in few sec on Bing images search ( type GTX580 slides ).


I add the white square, less disturbing.

(just have to redo the slide with same color sheme and update the game with new ( Dirt2>Dirt3, BC2 > BF3 etc ), or just mod it.


1 alone it was ok, but not both at same time.. ( its still possible Nvidia will use the old chart and moddifiying it, but it is really doubtfull, the 550 backround image slide you can easy find on the web, he's used for the spec )

Note: OB .... have remove his article and someone have too get this screen and discussed it before http://translate.google.com/translat...icherer-quelle

[trinibwoy]

or they are doing it knowing they earn less per card, but also undermine the competition and forcing them to earn less per card.

nVidia has no incentive at all to lower prices just to hurt AMD. That buys them nothing. If GK104 prices are low it simply means there's another chip coming to occupy the $350-$700 range.

[Handrox]

nVidia GK104
Product name (top solution): GeForce GTX 680 (but not entirely sure)
28nm production at TSMC, the area approximately 340mm ² (unofficial estimate)
4 Graphics Processing Cluster (GPC)
4 Streaming Multiprocessors (SM) aka shader clusters per GPC, ergo a total of 16 shader clusters for the GK104-chip
96 stream processors (SP) aka shader shader units per cluster, ergo a total of 1536 shader units for the GK104-chip
8 texture units (TMU) aka shader texture units per cluster, ergo a total of 128 units of textures for the GK104-chip
32 raster operation units (ROPs)
256-bit DDR memory interface (up to GDDR5)
Chip clock (top model): 950 MHz
Elimination of Hotclocks, no extra clock speed of the shader units more
Single-precision arithmetic performance 2.9 teraflops , double precision with 1:6 = 486 DP GFlops processing power, 121 Texturierleistung Gtex / sec
Memory clock (top model): 2500 MHz, memory bandwidth so that at 160 GB / sec
2048 MB of GDDR5 memory configuration

http://www.3dcenter.org/news/die-akt...rformance-chip

[Vardant]

Now where have I seen that before?

[Lanek]

chiphell maybe ? ( outside TMU count, clock speed, DP Flops etc )

Anyway, this bring an interesting question.. 4SM x 96SP bring 1GPC = 384SP (4gpc )... the exact count of the GF114 ( or 104 with 1SM disabled ).. or if you like better a 580 with 96SP /SM .
I ask me how performance wise work thoses Cudacores, cause it is clear they are " a lot smaller ", and this is not due to 28nm.. Outside other part have been drastically reduced in size or completely removed ( Polymorph engine, etc etc )

If Nvidia is going to something pretty similar to AMD this round ( SP wise ), this could be really interessant.

[Manicdan]

nVidia has no incentive at all to lower prices just to hurt AMD. That buys them nothing. If GK104 prices are low it simply means there's another chip coming to occupy the $350-$700 range.

it does get them something. if most of your profits come from server side, and your competition makes most from client side. you might try to sell your client side stuff as low as possible and making them less sufficient to compete due to lower margins and lets you gain more of their market share. its basically an oligopoly and the only sales you get are ones your competition did not get.

[Nightcover]

This might be relevant:
Rambus

[Vardant]

I don't think Rambus has anything to offer at this point, it was merely about stopping their patent trolling attempts.

[Lanek]

I don't think Rambus has anything to offer at this point, it was merely about stopping their patent trolling attempts.

exactly ... note, i dont know how much it have cost them, but surely a nice ammount for stop this.

[Vardant]

Maybe, or maybe it was "cheap", because Rambus recently lost a lot of its value.

[trinibwoy]

it does get them something. if most of your profits come from server side, and your competition makes most from client side. you might try to sell your client side stuff as low as possible and making them less sufficient to compete due to lower margins and lets you gain more of their market share. its basically an oligopoly and the only sales you get are ones your competition did not get.

You're making a very basic assumption that companies strive for monopoly status as a primary goal. A race to the bottom does nothing for nVidia's share price so no, there will be no freebies from the green team just to make AMD bleed.

If nVidia has an ace with Kepler they will use it to make more money, not less (see G80). There's also the question of supply - you can only undercut prices so much until you can't meet demand.

[AliG]

You're making a very basic assumption that companies strive for monopoly status as a primary goal. A race to the bottom does nothing for nVidia's share price so no, there will be no freebies from the green team just to make AMD bleed.

If nVidia has an ace with Kepler they will use it to make more money, not less (see G80). There's also the question of supply - you can only undercut prices so much until you can't meet demand.

I wouldn't say nothing. If they can sell enough units at a price that hurts AMD's sales, it would increase their market share and ergo their stock price. It's all about bottom line profits, not how you get there (think state functions). If they increase their bottom line while reducing their competition, then naturally more people will be willing to invest their money with the green team.

[trinibwoy]

There's no reason to believe nVidia can gain market share via lower pricing. AMD will simply match them. Net result is no market share gains and both companies lose. The company with lower costs simply loses less - only the customer wins.

Price wars work if one company has significantly lower costs due to economies of scale, vertical integration, operating efficiency etc. Otherwise it's a useless tactic.

[aussie-revhead]

Regardless of whether gk104 is top chip or not, if it competes with and possibly beats (in specific games) the 7970 then what stops Nvidia charging prices similar to 7970, or a little lower to force a drop from AMD? If its not the highest gpu then why would it be called a 680 and not a 670? The $299 price point makes it more like a 660 even ... There is no way any 680 labelled card will release at $299rrp, how much did 280-480-580 release at?

[Lanek]

Maybe cause this card have not the performance some want to make believe it have, and nothing is ready for the high end yet in the side of Nvidia ? ( pure supposition, im drunk and dont want enter a ranged war now .. )

The best i have read thoses last weeks, is peoples who think Nvidia want to lower the price of the market for going back, to a lower price for high end ... yes ofc..

[Handrox]

Sources are now telling SemiAccurate that Nvidia has two variants of the GK104 in the pipe. These two variants hint at a finer grained fusing ability for the end product.
The two siblings are said to be GK104-400 and GK104-335, basically a full working and partially fused off version of the same chip. The -400 is said to be an “8 group” device, the -335 described as “7 group’. If you recall the sad tale of Fermi/GF100, the chip had large swathes of shaders turned off, the ability to do less radical surgery was not there. This is a fairly painful way to deal with defects, the more granular you can make the disabling, the better off you are.
Nothing comes for free in the silicon world, and the art of chip design is balancing granularity with cost. Nvidia botched this badly in Fermi, and paid a high price. The only good that came of it was the entertainment in seeing their spokespeople spin ever increasing leaps of logic in public. This however doesn’t placate investors much, even if they do smile.
With this new description of the -400 and -335 variant of GK104, it looks like Nvidia has implemented what AMD has been doing since at least the R700 (HD4000) chips, if not earlier. Instead of being forced to fuse off large blocks of shaders as a minimum, it looks like they can now do much smaller chunks. In Fermi terms, instead of taking a CU at a whack, they can now do portions of a CU too.
This should greatly improve yields, allow for endless SKU variations, and generally make things better. Of course, it comes at a die size penalty, but after the last learning experience, it would be foolish to do otherwise.S|A

http://semiaccurate.com/2012/02/09/t...pler-variants/