Kepler Nvidia GeForce GTX 780

[DarthShader]

I play it by common sense and some well educated guessing.

Unfortunatly there is only wishful thinking in your post.

You are basicly saying, compared to a big Fermi, the mid Kepler GK104 drops die size by about a thrid ( ca. 510mm^2 is 50% more than ca. 340m^2), drops TDP by around a thrid (ca. 270W real peak TDP is 50% more than ca. 180W real peak TDP), increases transistor amount only a little bit (The rumours mention 3.5 Billions trannies, maybe up to 4. Billion) and the performance goes up by 50% too??? I can only see three possibilities how this could happen:
1) Fermi is the least efficient chip ever. Hot, broken, unfixable. Kepler is a heavily reworked architecture, picking up some low hanging fruits.
2) nV engeneers put physics and TSMC engeeners to shame, by doubling the gains from going to 28nm process TSMC would ever admit were possible.
3) What you say is wrong.

So what is going on, if there is indeed 3x times more Cuda Cores? Where does that 50% more perf come from? Dropping hot clock looks almost certain by now, but there seems to be still a lot more raw GFLOPs available. Will they translate to more performance?

It depends. My educated guess is ... there will be no SFUs anymore in a SM. That's were the space will come from, to fit all the extra CCs. Special functions will be done on the CCs in multiple clock cycles, just like... in GCN! There are a few advantages from this approach. First, you can reduce the data movement inside a SM, registers can be kept closer to the SIMDs. Data moving is expensive, so you save power by avoiding that. Furthermore, SFUs do nothing for linpack numbers, they don't increase the FLOP count. And nVidia promised to deliver 3 times more GFLOP/Watt with Kepler and HPC is a very important market for them. So if you exchange "useless" SFUs for shaders, saving some power by doing that, this goal becomes possible to achieve!

So if you look at artificial, "canned" benchmarks that rely on raw GFLOp power... yes, there is going to be 50% more performance. But if you look at others, ones requiring special functions... performance will start to tank, likely under the performance of a GTX580. Games require a mix of both, so who knows how it will balance itself out. Overall faster than a GTX580 is very likely IMO, but not by much.

[RPGWiZaRD]

I play it by common sense and some well educated guessing.

Well if we're going by common sense and educated guessing :P mine would be (without going into the boring details):

- GTX580, GTX570 will be EOL ASAP
- GK104 on avg 10~15% faster (will mostly range 5~20% faster than GTX580)
- GK104 was never meant to be a highend chip but due to AMD's not overly impressive performance for a new gen, Nvidia can now market is as that and can instead delay highend chip to try squeesh as much lower TDP as possible as that's most likely the reason it's delayed)
- GK104 was targeted to be around 300~$350 MSRP but due to AMD's high pricing of HD7950/HD7970 can place it at up to $399 and still have a very impressive performance/price ratio which is obviously needed to lure out customers into buying it for coming so late
- Noticably improved power consumption vs GTX580 (both beats it in performance and TDP)
- Will be a good card to buy from price/performance ratio at the time (due to coming months later than AMD, this is basic business 101)
- Most likely slightly better overclockability due to TDP headroom => overclockers can benefit a bit extra
- AMD's pricing will be dropped by $50 (max) or so for both cards
- What you get in comparision to Nvidia's last gen => save $100 ($150 comparing launch pricing) for ~15% better performance (slightly more for overclockers) vs lower power consumption (that you can expect reviews will point out like there's no tomorrow) + maybe some basic new feature we don't know about yet

TO ME this is reasonable. Not too much expectations, not too little, reasonable.

Further to reinforce my speculations, in case above would turn out to be true you can expect AMD's strategy was to bring out the cards as fast as possible and price them as high as possible as they knew a good chip from Nvidia was coming so they expected a drastic drop in sales while Kepler is out so therefore their strategy was to bring out the cards ASAP with as high price as possible to squeesh as much $$$ as possible from the months advantage they have to Kepler.

[ClockZ]

I think your on the right track if I were guessing^^

[Sentential]

Ya know what this reminds me of after looking at the die size? the 7900GTX. It wasnt as fast as the X1900XTX however it sure as sold a whole lot more. I think probably 5-10% faster than the GTX580 @ 250-300 is about what its going to shake out to be

[Tim]

50% would be awesome, even if canned benches. Especially if this would turn out to be the mid range Kepler.

On a side note, from memory I thought the 7900GTX was faster the X1900XTX. At least in a most games.

Seems like from this review in most of the games the 7900GTX is faster, think that was the general consensus? Pretty sure the X1900 was a whiner as well, whilst the 7900 had the Quadro cooler.

http://www.xbitlabs.com/articles/gra..._13.html#sect0

[Kuroimaho]

50% would be awesome, even if canned benches. Especially if this would turn out to be the mid range Kepler.

On a side note, from memory I thought the 7900GTX was faster the X1900XTX. At least in a most games.

Seems like from this review in most of the games the 7900GTX is faster, think that was the general consensus? Pretty sure the X1900 was a whiner as well, whilst the 7900 had the Quadro cooler.

http://www.xbitlabs.com/articles/gra..._13.html#sect0

The X1900XTX has much better minFPS, and that's what really matters, if everything was equal and the test was done right.

[Tim]

The only test I can find that shows minimum fps is the review from bit-tech where the xtx had HQ af turned on and it was (give or take) about even with the 7900GTX, I doubt HQ af made much of a difference. Would like to see otherwise though.

Anyway this is going way off topic lol.

[Kuroimaho]

The only test I can find that shows minimum fps is the review from bit-tech where the xtx had HQ af turned on and it was (give or take) about even with the 7900GTX, I doubt HQ af made much of a difference. Would like to see otherwise though.

Anyway this is going way off topic lol.

It's right in the bench you linked.

[Lanek]

I think there's a lot of misinformation or misleading between middle range and high end for Kepler.. and probably a lot of information, including the performance who have been mixed between sku number ( GK10x-gk104 ) and mddle range, top range... Basically if the card is 50-60% faster of the GTX580 in average, this is not the middle range and i doubt it have been designed for be this middle range. In this case this should not be the GTX560 replacement. ( or Nvidia have change their sku number scheme: GK100 > Quadro / tesla only aimed at full DP ratio, computing, GK104 > top level gaming....

This will bring the middle range card close of GTX580 SLI performance level... ( a bit lower, let say GTX590 level )...

Top range to top range, i can imagine without worry a +50% gain in average... But if the middle range top the GTX580 by 50-60%, this mean for middle range to middle range, the gain is of 90% easy ( 560 to GK"104" )... This will too said the low end or entry gamer will be capable of match or even be 10 % faster of a GTX580.. ( if the gk 104 is 50-60% faster of the GTX580, should we believe the GK106 can be 10% faster of a GTX580, 40-50% slower of a GK104? ) something dont match there...

Ofc, in some bench ( i think to Unigine for the example, assuming the gk104 should have same or more polymorph engine ( tesselation ), it couldend pretty higher of a 580 )

I think some source, insiders are using the term "Kepler"; and peoples write "GK104", then assume this is the 560TI replacement...

I dont even make a comparaison with the 7970 or AMD as you seen this is not the question.

[Kuroimaho]

I really wonder how could they even test these cards in china, why would nvidia release the proper drivers for the new cards.
So I won't take seriously any rumor which mentions performance and a Chinese source. Other details like ram, BW, they can tell but I doubt they can test the performance accurately.

[Tim]

It's right in the bench you linked.

D'oh lol. Thanks.

[GoldenTiger]

I really wonder how could they even test these cards in china, why would nvidia release the proper drivers for the new cards.
So I won't take seriously any rumor which mentions performance and a Chinese source. Other details like ram, BW, they can tell but I doubt they can test the performance accurately.

If anything, the performance would then be higher at proper release than with pre-release drivers . So, it's not something to dismiss.

[The Jesus]

I really wonder how could they even test these cards in china, why would nvidia release the proper drivers for the new cards.
So I won't take seriously any rumor which mentions performance and a Chinese source. Other details like ram, BW, they can tell but I doubt they can test the performance accurately.

They would be given test drivers specifically with the card/sample. Reviewers get cards days/weeks/months in advance to do all the testing. They had drivers obviously :/

A side note. Lanek, 50% over 580? I don't think it'll be that drastic. Somewhere around 30%, maybe 40 (tops). But this is a lot of speculation and nonsense that almost never holds true

[bamtan2]

we should expect it to perform close to a 7950/7970, and anything else is just crazy speculation

[Dimitriman]

50% faster than 580 with 180W and 350m2 die.. now I do want to see that happen. Count me in for purchase on sight.

But.. let's wait and see, not very long left now. But if you want my opinion, expecting that kind of performance from midrange is a pipe dream.

[cesariuth]

GK 107 (GT650M) with 384 CUDA cores tested on 3dmark 06
3dmark GK107.jpg

http://www.ozeros.com/2012/03/gk107-...-en-3dmark-06/

[Cold Fussion]

That's basically 8800gt levels of performance isn't it?

[highoctane]

That's basically 8800gt levels of performance isn't it?

It seems to be in that ball park range

Looking at my old 8800gt logs I got 14171
sm2.0-6560, sm3.0-6620, cpu-3162

That with a 6700gt running 9x400 8800gt sc at 675/987

[Andrea deluxe]

one day before GDC.......



funny situation!!!

[Tim]

Tehee that's going to be fun! Could be so fake tho, just something that is uploaded to googleusercontent it seems?

[Krizby87]

lol Tenerife, the place where 2 Boeing 747 collided head on, 500+ deceased, AMD must mean something here

[Vardant]

Looks like somebody is scared of GK104.

[HuffPCair]

lol Tenerife, the place where 2 Boeing 747 collided head on, 500+ deceased, AMD must mean something here

LOL I forgot about that, someone obviously didn't do their homework haha or they are just jerks

[Piledriver]

The post you replied to makes me really wonder how many posters are actually plants by marketing teams trying to divert attention from a competitors' product. But then again, there will always be someone who is "disappointed" or "waiting".

The fact of the matter is that AMD's next generation parts are nowhere near release. Think more along the lines of 2013 considering Tahiti refreshes haven't even taped out yet, let alone a move towards a more advanced architecture.

In addition, I think people have to be clear about one thing and I've been saying this from the start: The first generation of high end GCN parts were never meant to compete against Kepler. They were originally slated to be fabricated at 32nm and be a a direct competitor to NVIDIA's refreshed Fermi cards (GTX 580, GTX 570, etc.). Their performance results pretty much back this up as well.

IF (and that is a big IF) NVIDIA has designed Kepler to be a real step up from Fermi then AMD's current generation doesn't stand a chance in the high end market. On the other hand, AMD has proven that they can compete on price and all of their HD 7000-series (particularly in the $200+ price brackets) have more than enough padding to endure cuts of 10% or slightly higher.

This hilarious for some reasons.

First, one have to dig deeper to find a post where you don't praise\defend Nvidia and\or bash AMD. You calling someone out it's pure hypocrisy.

Second, you're so biased that someone commenting on how a GNC refresh could happen soon, made you go full tilt, making you miss the obvious, your “AMD’s marketing teams trying to divert attention from a competitors' product” is in fact a nvidiot, that only buys Nvidia but tries to be rational about it, by coming up with excuses like the well knowed “Let’s see what both have to offer” and was rationalizing again, by coming up with the excuse on how it acceptable to wait months for fermi\Kepler but it's not to wait months for the new AMD card, because one his head one was a refresh and other was a new arch and the guy that you quoted was in fact pointing the flaws on that reasoning is, because Kepler is as much of a new arch as the GNC refresh will be.

What a monumental epic fail from you.

Third, AMD is about to releases new card, a flood of news of how awesome Kepler is and how close it is, AMD is about to release other card, another flood of news of how awesome Kepler is and how close it is, Nvidia past years does nothing more than to try to stall the competition sales with diagonal schemes and propaganda for idiots. You have no issue with it. But the question is raised on why is makes sense to wait months for Nvidia, but it doesn't to wait for AMD, you take a issue on it, even if it was a nvidiot bringing it up.

if NVIDIA has designed Kepler to be a real step up from Fermi then AMD's current generation doesn't stand a chance in the high end market.

I’m sorry it’s been how many years in a row now that AMD have the fastest card? With comments like that that one, you sure can call someone out for doing marketing for someone.

Oh a pre-emptive strike on the moronic single-GPU argument. Matrox have the fastest card with a niche configuration for a niche market, and no one cares, just like no one but nvidia fanboys care about the single cpu nonsense. But hey if believing that AMD, if they wanted couldn't make a 600mm gpu also, makes you sleep better at night, whatever.

[Andrew LB]

lol Tenerife, the place where 2 Boeing 747 collided head on, 500+ deceased, AMD must mean something here

Reminds me of how AMD was going to "bulldoze" Intel with their latest CPU....