Printable View
You don't jump from 12m/mm2 to +14m/mm2 at the highend when you can't even hit 12m/mm2 for your lower chips.Quote:
Originally Posted by highoctane
If transistor count is right at 7.08b then they are looking at a best case of ~540mm2 to a worst case of ~585mm2.
It doesn't matter. If you go from 80 to 120, there is a 50% improvement, or a 33% worsening, depending on where you stand. But it doesn't matter if you are talking in %, mph, liters or whatever, math is the same.Quote:
Originally Posted by Handrox
---
ON TOPIC: I still don't understand how is it possible that a chip not designed for rendering can perform very close form a GTX690 whilst having a lower power consumption, on the same node. Unless the benchmarks used are cherrypicked it seems veeeery weird.
Regarding large dies and heatspreaders:
Guys, this may be a silly question and I'm sure there is a logical answer for this but what stops gpu manufacturers putting multiple cores in a single die - CPU style? Surely now, looking at how small and powerful the 680 is, this would become a possibility, right?
Counting down the days until Titan is released....
:D
~585mm2 would be pretty much be a 1:1 transistor avg scale up of gk104, I surely don't think that will be the case.Quote:
Originally Posted by LordEC911
If the chip has allot of cache for gpu compute it that will push the transistor count fast but should be laid out densely compared to logic.
Either way Nvidia is not new to making large dies, if anybody can pull off engineering an insanely large 7b transistor die for retail consumption it's Nvidia.
In some sense, that's essentially what CUDA cores are. :p:Quote:
Originally Posted by aussie-revhead
Quote:
Originally Posted by LordEC911
SourceQuote:
Originally Posted by fellix@B3D
Yeah even a display port would work, BUT i use a Apple cinema display with my GTX 590, which only has a male mini display port on it so I need a Male display port to Female mini-display port... which I cant find ANYWHERE.Quote:
Originally Posted by zalbard
So Female mini display to Male Display port...
Somebody please tell me what the hell this cards supposed to be. I havent been up to date for the last 5 or 6 months + did not read all the pages and am now unsure about its place/rank/...
So... is this "TITAN" the flagship single-GPU of the next generation (780GTX?), with lower spec cards following? Or is this just THE Kepler, that was replaced by GK104 at launch?
What confuses me the most is the price (900usd?) and the delay of AMDs next gen cards.
Thanks in advance.
Cheers
It's just the big Kepler that was supposed to come back when the GTX 680 was originally released. No new gen is coming any time soon, though this will be able to compete with that too. Most likely though this will stay a curiosity like 8800 ultra back in the day, mostly just for PR and some xtreme enthusiasts, and will never actually drop in price to reasonable levels.
Estimatiom of 520mm2 ... not 502mm2 ... as said 14.1M of transistors / mm2 will be really surprising on the 28nm2 ... and like the GK 110 who will be use on Titan have been allready released 6 months ago with the Tesla K20...Quote:
Originally Posted by zalbard
?If you believe the GK110 = GK100 you are really far of reality . The Power of the marketingQuote:
Originally Posted by Pantsu
I don't think gk100 could have been released at gk110 speeds, but I wouldn't doubt if Nvidia could have released a cut down lower clocked part that was 25% faster than a gtx 680. However it would only serve to hurt Nvidia financially as such a card would reduce the price of the gtx 680(as it couldn't be called this anymore) and the super low number/high priced nature of gk100 would hardly offset the difference. 28nm seems to be less troublesome for both companies this generation and I can imagine yields are better(not supply of wafers). However financially 28nm isn't as good for Nvidia as 40nm, as they have to pay for the wafers rather than the number of good dies so monoliths makes less sense to produce for mass production. Its best to wait for yields to be good I can imagine and sell the early stuff as high a margins as possible in the professional market.Quote:
Originally Posted by Lanek
When Nvidia turned the a gtx 660 ti(very likely) into a gtx 680, while naming gk110 into geforce Titan. they showed a very intelligent marketing move. By bumping up a efficient gaming card that doesn't consume massive amounts of power at high voltage, Nvidia made the gtx 680 look like a better card than it really was for 500. With the absolute nutty pricing of the 7970, it made it really look like Nvidia took home every metric of power, performance and price.
Nvidia took advantage of AMD's idiocy(low clocks, high price) and made it so easy to make the gtx 680 look good.
The gtx 680 isn't really better than a 7970, however AMD's blundering marketing team made it possible for a overclocked gtx 660 ti look like a darling flagship card.
Who gives a fluke about GK100 ? GK110 was taped out before GK104 was on the street.
GTX 690 was from start limited edition, i am not sure, but they have only 2000 pcsQuote:
Originally Posted by Dimitriman
Bubble> Launch is in six days, yes six not five. This is the most powerfull VGA in history, only few percents under dual chip GTX 690. Without microstuttering, game profiles, noisy cooler, but for dual chip price. Price is the same as GTX 690. Card is really beautiful, looking really great. Cooler was made with great materials, like GTX 690.
Well that's two false statements...Quote:
Originally Posted by PedantOne
OBR removed the results of your website, reason?
Oui, I really hope this isn't just a rumor and this this don't end up coming out.
Because I just sold my 680 lightning for $350...
NEO- I would have bought your Lightning for that .... used ones are still $500+ here
Lets hope for miracles...Quote:
Originally Posted by Lanek
Im sorry, I suspected this was the original big kepler also, whats the diff between GK100 and GK110? (excuse the n00b question...I just bench them...)Quote:
Originally Posted by Lanek
:D
There is/was no GK100, at least not in any conceivable form. Drawing boards maybe.
You're the one really far off reality if you think that's what I said.Quote:
Originally Posted by Lanek
LOL The "680 is really a 660!" debate will never end. As far as I know, cpu and gpu manufacturers always have multiple designs on the drawing board and decisions on what to sell are governed by what can be sold most profitably. So if GK100 or whatever couldn't be sold 11 months ago as a product line, what turned out to be the 680 was. Similarly, AMD probably had much higher clocks intended for the 7970, or perhaps even hoped to have the 8000 chip done by then, but we got the 7970 because that's what they could produce at that point in time.
Everyone seems to forget these are for profit companies whose primary concern is to make money for shareholders, not give you cards that you can overclock the highest, or cards that perform the highest if doing so comes at a loss on current process.
There were rumors NV would put out a mid range chip that offered 90% of the performance of the 7970 before the 680 launched, and that a bigger chip would come in the fall. At the time I said,"If they price it $100 less it will clean up based on branding, and feature set". Imagine their delight when it turned out 110% of the 7970.
Quote:
Originally Posted by Lanek
Exactly what boxleitnerb said.Quote:
Originally Posted by boxleitnerb
There is no GK100, never was.
Nothing new :)Quote:
Originally Posted by Cordell Cahill
Corell> Everything from that article is known many days, Titan will not beat GTX 690, but its a single GPU against bloody problematic dual GPU
Damn,I must have missed it while trying to skim over all the back and forth bickering in hereQuote:
Originally Posted by PedantOne