NVIDIA promises to ship Kepler this year

[BFG10K]

Of course, DP performance / watt doesn't necessarily directly translate into a similar gaming performance increase. Still, sounds like Kepler will be a beast.

DP performance / watt is an irrelevant metric for games given they don't use DP.

 

[apoppin]

That could be a new card to review like a 6870x2 or something else.
At best it could be a mid life kicker card like the 4770 was. That would be nice, but I doubt it. I'm sure we would have heard something if a next generation card was about to be reviewed for release in the next 2 weeks.

i get dragged back in
:$

It's a new partner card

 

[notty22]

please google the term 'paper launch nvidia' and you will find the track record i was referring to.

i wont say that the other camp is innocent, though.

You can do the same thing, like this.
Google ATI paper launch
Google AMD paper launch
You get over 30,000 hits for each.
[Microprocessor Forum] ATI's "paper launch" ridiculed


I know you said either was innocent, I'm just refreshing it for others.

 

[happy medium]

i get dragged back in
:$

It's a new partner card

As in new vendor?

 

[mnewsham]

As in new vendor?

I read it as XFX or MSI is putting out a new version of a card (special OC edition or something like that) But that is only how i read it. Could be wrong

 

[Ares1214]

Hmmm, right, Ill believe this when I see it. I have a feeling if they moved it up, its gonna be a pretty shaky process and they are rushing it to compete with AMD, and if they are lying, we may not see these cards retail till Q1 or Q2 2012. AMD of course also isnt exactly always on time with their GPU's, but lately they have been doing far better than NV in that area. If NV says this is coming out before the end of the year, we wont get one until X-mas or Valentine's Day, if AMD says they are shipping before the end of the year, I'd expect more of a Q3 to Q4 release. However the 7xxx series also might not be as impressive as Kepler, seems like AMD is more interested in their 8xxx series and the whole SIMD arch change.

 

[RussianSensation]

DP performance / watt is an irrelevant metric for games given they don't use DP.

:thumbsup:

That's true. Still, they likely increased DP performance by having a much more complex chip though. But like you said, that doesn't tell us how much faster it will be for games.

 

[apoppin]

As in new vendor?

No. AMD offered me a few new partner cards to pick from. They have really matured the 6000 series with some pretty unique product and the choices are really tough to pick one.

However, i am interested what Gigabyte does with passive cooling on a fairly fast card and asked for a 6770
- i hope AMD sends ABT that card. It intrigues me since they also talk about overclocking it.

And my own feeling is that AMD and Nvidia will launch within a month of each other

 

[Genx87]

if nvidia's newest top card gives like ~15% higher performance than gtx 580 and 1/3 the power, they may have a problem if AMD brings out a card that is 2/3 highter in performance than 6970 and at about 15% less power.

Nvidia's track record is to match the previous generations dual GPU card with a single GPU. Look at the 590GTX. That is where I expect Kepler to be performance wise in games. In compute I wouldnt be surprised if it kicks the 590 in the nutz!

 

[SirPauly]

No. AMD offered me a few new partner cards to pick from. They have really matured the 6000 series with some pretty unique product and the choices are really tough to pick one.

However, i am interested what Gigabyte does with passive cooling on a fairly fast card and asked for a 6770
- i hope AMD sends ABT that card. It intrigues me since they also talk about overclocking it.

And my own feeling is that AMD and Nvidia will launch within a month of each other

There really hasn't been clarity on AMD's point-of-view and vaguely offer 7XXX series, where nVidia mentions Kepler. For all I know, this means AMD may release a 7XXX series in a mid-range chip that offers similar performance as their current sku's but may offer improved performance per watt and size -- let's say a replacement for Juniper which was the 5770/6770 -- and then over time offer other family members.

 

[apoppin]

There really hasn't been clarity on AMD's point-of-view and vaguely offer 7XXX series, where nVidia mentions Kepler. For all I know, this means AMD may release a 7XXX series in a mid-range chip that offers similar performance as their current sku's but may offer improved performance per watt and size -- let's say a replacement for Juniper which was the 5770/6770 -- and then over time offer other family members.

i thought that IS what we expected. That or an even lower-end card to test 28nm first.

This coming next-generation is all about Perf per Watt, i would project. Performance increases will be significant but not like the next-next gen (700/8000 series)

And now i gotta go. i am seriously under a fast-approaching deadline. See you next week (here).
Aloha

 

[Jionix]

This strikes me as nothing more then a ploy by Nvidia to potentially eat up the limited 28nm lines from TSMC with their initial terrible yields. This is a strategy with proven results for them.

TSMC has said they have cut 28nm lines for Q4 because of "lack of orders". 3% of their overall production will go to 28nm in Q4. This is still a potentially large amount of chips for any one company --- ~500,000? Depends on size of chip and how f'ed-up 28nm is.

Either it's complete BS from Nvidia, because just a month ago they had no interest in launching in 2011, or they have uncovered something about AMD's plans (maybe a source on performance), and now they are going to plan B to ruin the party.

 

[Saico]

As AMD stated, they are going to release GCN cards this year. I assume that GCN is for high-end only, because mid-range cards should be vliw-4 to crossfire with trinity.

So, I'm expecting 7970 in October.

 

[bryanW1995]

"Start shipping" does not at all mean what it implies.

I think most people will read those words and think that means sometime in Q4 they will be able to order a Kepler-based GPU from Newegg.

But "start shipping" could literally mean nothing more than sending out chips to AIB's, not that those AIB's will even receive them in Q4 because Nvidia didn't put them into the mail until Dec 31st.

The thing the public cares about is when they can order the product and have it delivered, not when Nvidia is achieving yet another milestone that must occur in a long progression of other milestone events that all have to transpire in order for the product to show up at Newegg.

If they are merely going to "start shipping" by the end of the year then I will be impressed if I can order one from Newegg before Q1 ends. I won't be surprised at all if the reality is that I can order no sooner than April (Q2).

That's what "start shipping" means to me, I can expect to be able to buy it in early Q2. Kudos Nvidia, job well done, pats on the back all around :thumbsup: This is different somehow from how Fermi went down?

I think that it's way too early to start comparing kepler to fermi. Fermi I failed not just because it was late, but because it was 7 months later than AMD. Also, Fermi I was too hot/loud, but again that was just a relative. Outstanding performance could have saved it, but 5970 was still much faster and not any more hot/loud than gtx 480. This round we've heard lots of talk from amd but no hard release date yet, and we have no idea how kepler will compare performance-wise with SI. IF southern islands releases in september and is very close to the eventual kepler in performance + noise/heat/power AND kepler comes out in april then we'll have a repeat. However, remember that fermi II beat amd to the punch and improved by leaps and bounds in noise/heat/power + it ended up significantly faster than amd's refresh. Admittedly amd wasn't 7 months behind, but there sure were a lot of extremely disappointed red-teamers in mid december last year.

 

[bryanW1995]

No. AMD offered me a few new partner cards to pick from. They have really matured the 6000 series with some pretty unique product and the choices are really tough to pick one.

However, i am interested what Gigabyte does with passive cooling on a fairly fast card and asked for a 6770
- i hope AMD sends ABT that card. It intrigues me since they also talk about overclocking it.

And my own feeling is that AMD and Nvidia will launch within a month of each other

Is that a "feeling" based upon things that you've seen/heard, or is it based upon conversations with reps from amd and nvidia?

I'm interested to check out a fanless 6770. Maybe that could replace my fanless 7300gt as my backup/emergency card

 

[SirPauly]

I think that it's way too early to start comparing kepler to fermi. Fermi I failed not just because it was late, but because it was 7 months later than AMD. Also, Fermi I was too hot/loud, but again that was just a relative. Outstanding performance could have saved it, but 5970 was still much faster and not any more hot/loud than gtx 480. This round we've heard lots of talk from amd but no hard release date yet, and we have no idea how kepler will compare performance-wise with SI. IF southern islands releases in september and is very close to the eventual kepler in performance + noise/heat/power AND kepler comes out in april then we'll have a repeat. However, remember that fermi II beat amd to the punch and improved by leaps and bounds in noise/heat/power + it ended up significantly faster than amd's refresh. Admittedly amd wasn't 7 months behind, but there sure were a lot of extremely disappointed red-teamers in mid december last year.

I just don't understand this thinking. Fermi didn't fail and is bringing in impressive revenue, margins, profit, market share and did bring impressive innovation, features, strengths, value, choice and competition.

Fermi stumbled in the beginning based on execution but if Fermi failed than AMD has failed much worse because they're losing discrete share to a failure, based on your thinking.

 

[Dark Shroud]

it ended up significantly faster than amd's refresh. Admittedly amd wasn't 7 months behind, but there sure were a lot of extremely disappointed red-teamers in mid december last year.

What do you mean by this part right here? The GTX 590 was a failure and the GTX 580 does not destroy the HD 6970. Roughly 10% more power for over $100 more in price. It's a similar story for the GTX 570 vs the HD 6950.

 

[Seero]

My uneducational guess on 40nm is some layouts that were used to work on 45nm can no longer be made, but they didn't know ahead so the yield rate plummets. With this failure, TSMC and others should have used a lesson and specify what layouts can be make, and what can not be used on 28nm. That means both GPU makers will design their 28nm model correctly to improve yield rate.

Assuming my uneducational guess is true, than most of the 4xx and 5xxx are work arounds, so it should be relatively easier to improve performance on 28nm as it is smaller, and the design will be much cleaner. That means, if true, then the next gen from both comp are going to rock.

Please correct me if i am mistaken.

 

[Arkadrel]

December 31st?

QFT

This means they ll start shipping cards around the world, on December 31st.
It also means that you ll be able to BUY the cards 2months lateron (in 2012).

My guess?

Able to buy Nvidia's 6xx series around Feb 2012.
Able to buy AMD 7xxx series around Sept. 2011.

In short Nvidia still end up being around 6 months behinde AMD.

 

[96Firebird]

but there sure were a lot of extremely disappointed red-teamers in mid december last year.

And how quick they are to forget. But any delay by Nvidia is a tragedy that will kill the company.

Really, there is no date set for either company. So all these predictions on release dates are just hearsay.

I will wait for releases by both companies and pick the one with best performance/$, with some overclocking abilities as well. Can't wait to see how close they clock these chips to their max, being a new process it is kinda up in the air.

 

[StrangerGuy]

I'll take this with the Southern Islands announcement as an indication that TSMC's 28nm isn't as crappy as their horrendous 40nm node.

 

[bryanW1995]

I just don't understand this thinking. Fermi didn't fail and is bringing in impressive revenue, margins, profit, market share and did bring impressive innovation, features, strengths, value, choice and competition.

Fermi stumbled in the beginning based on execution but if Fermi failed than AMD has failed much worse because they're losing discrete share to a failure, based on your thinking.

What do you mean by this part right here? The GTX 590 was a failure and the GTX 580 does not destroy the HD 6970. Roughly 10% more power for over $100 more in price. It's a similar story for the GTX 570 vs the HD 6950.

Do I really need to go back and dig up all of those 69x0 launch threads? There were a LOT of wild speculations in there about how great the 6970 was going to be, how it would trash gtx 580, etc etc etc. The silence was deafening after the reviews actually came out. And my point was more that gtx 470/480 were disappointments. And that gtx 570/580 were not a disappointment, largely due to a more timely delivery of part 2 and a significant improvement in noise/heat/power vs Fermi I.

I do find it humorous that I'm being attacked for being too pro-AMD and too pro-Nvidia in consecutive posts.

 

[SirPauly]

Because it is extremism on your part by offering Fermi failure and an extremely disappointed red team.

Fermi is an architecture, which many product sku's were based on. And don't remember an extremely disappointed red team either.

 

[sawtx]

GTC 2011 is in October, so maybe March/2012 ?

Note that GTC got delayed, it is now in May 2012.

 

[SirPauly]

GTC in the United States was moved back but had their first GTC in Japan:

http://blogs.nvidia.com/2011/07/gpu-technology-conference-1000-strong-in-tokyo/