Sign in with
Sign up | Sign in
Your question
Closed

Nvidia's Kepler to Arrive in 2012

Last response: in News comments
Share
Anonymous
a b U Graphics card
August 5, 2011 11:44:31 PM

At least that means my current one will live longer...
Score
10
Anonymous
a b U Graphics card
August 5, 2011 11:45:39 PM

Moore's law is already slowing down, it's only going to get harder to design these ridiculously big and hot chips that Nvidia loves so much.
Score
21
Related resources
a b U Graphics card
August 5, 2011 11:45:42 PM

good-luck, I'm waiting for you and you better come correct.
I'm hating all these AMD/Radeon fanbois talking all this nonsense about the HD 7xxx series.
(nVidia: the way it's meant to be played...)
Score
-14
August 5, 2011 11:57:55 PM

"According to information released by Nvidia so far, Kepler cards will triple the dual-precision floating point performance of Fermi and hit up to 6 dp GFlops, while its successor Maxwell (scheduled for a 2013 release) is expected to with almost 16 dp GFlops. These are big promises and Nvidia wouldn't want to miss them."

...sigh,

You left out a small yet crucial piece of information that completely changes the meaning of this last paragraph. Per W. Nvidia says Kepler will triple the double precision performance per Watt over Fermi, they've never said anything about a flat out tripling in dp performance.
Score
27
August 5, 2011 11:59:17 PM

How about some proofreading and checking your facts...6 (or even 16) dp gflops is way below most GPUs and many CPUs. Do you mean GFLOPs/watt as the chart that's made the rounds says? That's a little more impressive.

/Thinking ahead to Maxwell or post-SI as next upgrade cycle.
Score
19
a c 173 U Graphics card
a b Î Nvidia
August 6, 2011 12:27:03 AM

This wasn't up when I made a post about this subject this morning. I'll copy what I wrote in this thread. Props to dragonsqrrl who caught what I did and Charlie missed.

http://semiaccurate.com/2011/07/05/nvidias-kepler-comes...

I know you guys hate him, but he was correct with Fermi. Anand is also running an article saying the same.

http://www.anandtech.com/show/4572/kepler-gpus-shipping...

Quote:
Back to the article on Fudzilla, the second paragraph contains three bullet points, that the chip has taped out, there is a lot of leakage yet to be dealt with, and the 40nm to 28nm transition is tough.


Remember that for each respin of silicon, you'll need to tack on another 6 weeks or so. If tape out happened in June, the first respin won't be done until mid July at the earliest, probably more like August sometime. IF that one is the one they want to run with, production is 3mo, so they could have something ready by Nov. Nvidia has already said it's not happening, so the leakage much be bad.

I'd like to point out there seems to be an issue with the S/A article. He keeps going on about 2.5x the number of transistors, but I think he missed something. S/A quote.

Quote:
First is that Kepler has about 2.5x the DP floating point performance of Fermi, at least according to Nvidia projections.


That doesn't quite match what Anand reported.

Quote:
For Kepler NVIDIA is expecting "about 3x improvement in [double precision] performance per watt"


I added the bold part as I think this is what Nvidia meant. I think Charlie missed that. Other then that error, I don't think you'll be seeing Kepler for sale until H1 2012. Right now it depends on how many respins they need to fix their leakage problem.
Score
0
August 6, 2011 1:45:12 AM

figures
Score
-1
a b U Graphics card
August 6, 2011 4:47:55 AM

Finally!
i hope the cards are good, might upgrade my aging 9600GT finally...i just hope a smaller architecture means less heat, and consume less power...

Hope they're cheaper than fermi at launch ;) 
Score
3
August 6, 2011 4:56:45 AM

I for one am happy to see that a hardware company is actually SLOWING down the release of a product, because so many times in recent history product gets rushed out the door and is not completely tested and or checked as thoroughly as it should have been.

I'd rather wait a bit longer and get good durability, usability, and reliability than get a product that was rushed to market plagued with problems.
Score
6
August 6, 2011 5:30:52 AM

4745454bThis wasn't up when I made a post about this subject this morning. I'll copy what I wrote in this thread. Props to dragonsqrrl who caught what I did and Charlie... I added the bold part as I think this is what Nvidia meant. I think Charlie missed that. Other then that error, I don't think you'll be seeing Kepler for sale until H1 2012. Right now it depends on how many respins they need to fix their leakage problem.

I'd expect nothing less out of Demerjian...
Score
-5
a b U Graphics card
August 6, 2011 10:58:28 AM

(god I love when I post something when Im not logged in then log in on the pop up so what I posted will post and it doesnt post... thats great. Lets have some more of that!)

a good article on the issues that tsmc is having...

http://news.softpedia.com/news/AMD-Southern-Islands-and...
Score
7
a c 173 U Graphics card
a b Î Nvidia
August 6, 2011 12:03:57 PM

Quote:
Finally!
i hope the cards are good, might upgrade my aging 9600GT finally...i just hope a smaller architecture means less heat, and consume less power...


What are you smoking? Who said anything about a smaller arch? I'm sure Nvidia is piling more shaders into their chip, so even with the die shrink I bet it will be larger then Fermi.

Quote:
I'd expect nothing less out of Demerjian...


Sometimes your hatred of a company means you miss things. Good thing I don't hate Nvidia, just dislike them:p 

Quote:
a good article on the issues that tsmc is having...

http://news.softpedia.com/news/AMD [...] 2552.shtml


Interesting, though I'm not sure it explains the issues TSMC is having. I wonder after reading that however if this means AMD will have clock speed with their cards. Low power usually equals lower clocks as well. I wonder how this will impact AMD.
Score
-5
August 6, 2011 12:21:02 PM

Quote:
Quote :
a good article on the issues that tsmc is having...

http://news.softpedia.com/news/AMD [...] 2552.shtml


Interesting, though I'm not sure it explains the issues TSMC is having. I wonder after reading that however if this means AMD will have clock speed with their cards. Low power usually equals lower clocks as well. I wonder how this will impact AMD.


It depend on how long time it will take Nvidia to get Kepler right with more advanced technology. If they are lucky. Nvidia has much better GPU this time. If they are unlucky, the AMD is ready for their 8xxxx series when TMSC finally manage to make desent chips to Nvidia...
Score
2
a c 173 U Graphics card
a b Î Nvidia
August 6, 2011 1:10:29 PM

I was thinking about this over my last hour of work. It might not be bad for AMD. The article said something about AMD using this low power process for its 7670 and 7650 chips? These would be the upgraded 6670? AMD could do this to be "first", while waiting for the HKMG 28nm process to get finished and then release the rest of their chips on that. If anyone has any info on what the clock speeds could be on the low power process I'd love to see it.
Score
3
August 6, 2011 1:15:24 PM

So basically hold on to your current card till 2013. Doesn't make sense to buy a card during 2012, then have its big brother come out a year later. As slow as the gaming industry is I doubt I could justify replacing my current card for another 3 years even.
Score
-6
August 6, 2011 4:46:57 PM

I think the reason things are seeming to slow down is it is getting harder and harder to push silicon any smaller. Pretty soon we are going to have to switch to different materials, there's a limit to what silicon can do, and it seems we are reaching that within a decade or so
Score
4
August 6, 2011 7:48:22 PM

ch1pluvrMoore's law is already slowing down, it's only going to get harder to design these ridiculously big and hot chips that Nvidia loves so much.

They don't use that much power when idle. I like new video chips, I can use them to encode video faster and to run folding@home/seti@home for science. Plus it makes all the cooler running chipsets cheaper.
Score
1
Anonymous
a b U Graphics card
August 7, 2011 1:32:07 AM

danwat: Been drinking? I fail to see how that applies to the engineering difficulties associated with the next Nvidia chip being the size of a dinner plate.

Besides, anything you've described could be done with a much more reasonable GPU like a Radeon 6850 or below....
Score
0
August 7, 2011 3:13:09 AM

nVidia should have used Gallium-Arsenide compound semiconductors for 6xx series.

/thread

owait, GaAs contains extraordinarily high levels of impurities in relation to silicon and thus cannot be manufactured past a 500 nm node, not to mention gallium is rarer than gold and arsenic is very toxic and a carcinogen.

nvm

nVidia, you do good job, keep going with your Kepler stuff.

(P.S. GaAs is just plain BA. The Cray 3 ran off of it, and it can support frequencies up to 250 GHz)
Score
1
August 7, 2011 11:31:21 PM

zxcvbnm44nVidia should have used Gallium-Arsenide compound semiconductors for 6xx series./threadowait, GaAs contains extraordinarily high levels of impurities in relation to silicon and thus cannot be manufactured past a 500 nm node, not to mention gallium is rarer than gold and arsenic is very toxic and a carcinogen.nvmnVidia, you do good job, keep going with your Kepler stuff.(P.S. GaAs is just plain BA. The Cray 3 ran off of it, and it can support frequencies up to 250 GHz)

LOL. Just needed to throw that in there.

As for this delay, I'm not really surprised. Nvidia is starting to try to change the game up a bit and will encounter problems along the way, die size changes are expected at this level. I personally prefer AMD, but, I hate seeing any company stagger. The better production and progress from each company (AMD, Intel, Nvidia), the better the prices for us. Competition is what drives the market. I'd just love to see programmers finally utilizing some of the hardware we already have, rather than see more upgrades.
Score
0
August 8, 2011 12:57:27 AM

So what about the AMD 7000 series?
Score
1
August 8, 2011 3:24:25 PM

hey maybe someone should point out this whole diamond transistor technology that you were talking about on here a few days ago to Nvidia.
Score
1
August 8, 2011 7:21:31 PM

Better to produce a fine product instead of rushing.
Score
2
September 25, 2011 5:16:51 AM

Maxwell will be crazy. I mean applications probably won't take advantage of it for a long time right?
Score
1
Anonymous
a b U Graphics card
October 20, 2011 5:54:45 AM

There aren't just technical limitations, but economic limitations as well. With this poor economy, the existing cards are not selling as fast and I'll bet they are really holding out so that the stock of existing cards will be sold during Christmas 2011. Businesses squeeze as much as they can out of one tech before rolling out the new one.
Score
1
October 21, 2011 8:01:40 PM

new003[/nomMaxwell will be crazy. I mean applications probably won't take advantage of it for a long time right?


a good ray tracing program could fill most of the extra processing power, and the rest could be used to add tessellation, true 3d rendering and depth of field...
theres a long list of things that they can improve right away to fill that void.
Score
1
October 21, 2011 8:09:12 PM

carbon nano tubes should be used to build gpus, lol they have superconductivity due to quantum tunneling which would mean that there power is only limited by the amount of energy you can supply, and they would have 0% heat generation.... nVidia please get away from silicon its already to small to have a reasonable loss of performance per nm.


ps. sorry if this is already posted but its not showing on my screen
Score
1
October 21, 2011 8:22:21 PM

let me clarify, that last statement nano tubes can be superconductors or semiconductors depending on there arrangement and the heat could be generated in the semiconductor state.
Score
1
October 26, 2011 3:26:58 AM

I predict stagnation at about 2020 due to atomic level limitations
Insulator needs to be three atomic layers thick
then
after awhile..the carbon nanotubes will allow for 1/1000 of wattage
and the 3D stacking game begins - a new boost in performance
but that's the limit folks - the future looks bright
but at the end of the tunnel is a mountain wall
unless...
someone invents subatomic particle quarkputers
which operate at zillions of ZetaHertz speed...
Score
0
November 20, 2011 1:58:34 PM

@CyberAngel

quantum computers can handle some image data in theory and since 1qbit per second = 1^2bit per second (500bit/s = 250000qbit/s) the resulting speed would be much faster. so there is definatly smaller and faster GPU beyond nano tubes... they just don't resemble todays computers...
Score
0
December 12, 2011 5:33:53 PM

I know its off topic a bit, but if you guys are trying to talk about a new way to build these cards... why havn't Intel, Amd, and Nvidia try and explore more in Quantum chips. i know the technology is off by a few decades yet. but it cant hurt any. and as of right now the only functioning quantum computer has about 60 sides on a single core. since its only an experimental rig yet can u imagine if it was not buggy. theoretically a chip, you could put 100s of sides on a core. and compute 10x faster than chips with the same amount of cores. yes one flaw would be power use and heating. but if you find a way to make the core be able to you less power. heat drops a lot. then refine the cores itself and you have something that could possibly be built in a computer sooner than u think. the first generations would not need to have some huge amount of sides just 4 in my opinion. and that would double if not triple the speed in itself. but that is just starters. hence binary would be obsolete at this point. which still wont be to far off from being obsolete at this point anyways at the rate we are moving in technology. Binary i personally give it about 10 years tops before everything goes to hexadecimal anyways. and another thing we would end up moving back to sign waves (analog) but instead of multiple ups and/or downs equal a 1/0 you could theoretically multiple spots of which a core can compute as a curtain hexadecimal letter or number. like in hexadecimal. so technically in one up down cycle you could have computer every single hexadecimal unit in half of a wave and the other half do the same thing... then along with that note all you could need is 16 sides on a single core to make such feat possible. and just adding more cores will only reduce the amount of lag it takes to go from side to side and compute the single digit. which i do know this is off topic of specifically graphics cards. but the cores could go from 512 in nvidias' 580 and 15xx in amds' 6970 you could have about 2 cores in itself and compute just as much. and that is including the amount for the ram on the chip. hence we now finally got a computer that will be able to run on hexadecimal without writing a code to make it into a binary number and compute that way. and hence you double if not more the speed of the item is being computed. with just that alone.

so the multiple cores. More than half the time computing hexadecimal is lost. the fact that it could compute two entire hexadecimal sectors in one single up/down pattern the single computer could be 16 times faster with 16 sides on a single core than a quad core right now with the 16 sided cores are . and that is is just the beginning of quantum mechanics.
Score
0
!