NVIDIA reveals Fermi's successor: Kepler at 28nm in 2011, Maxwell in 2013

Page 4 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

badb0y

Diamond Member
Feb 22, 2010
4,015
30
91
I'll beleive it when I see it.

I mean see it fuctioning not just in his hands.

;)
 

blastingcap

Diamond Member
Sep 16, 2010
6,654
5
76
Mabe he's not? At least I havent heard anything except a simple roadmap and performance targets, kinda like Intel does.

NV still hasn't rolled out the entire Fermi line, but NV's CEO is already talking about the future a lot these days rather than their present. From what I've read, Fermi will likely see a couple of more low-end derivatives and a dual-GPU card and then that's it. No refreshes--why bother, when 28nm is so close and they screwed up the architecture for 40nm anyway?

"If I don't have a present, I will talk about my future ... I've got a great new architecture coming and when it comes I will tell everybody about it. Until then, I let somebody else go talk smack. You don't hear Apple standing up on a podium and say we have great Macs coming because it is going to have this and this. They don't talk like that. Market leaders don't do that. You're not supposed to tell all your competitor's your secrets." -Jen-Hsun Huang, CEO of NVIDIA on June 16, 2009 http://www.brightsideofnews.com/new...uang-pulls-a-180-degree-turn-in-100-days.aspx

"For the very first time in the history of our company," Huang during his Tuesday keynote at the company's GPU Technical Conference in San Jose, California, "we are going to tell you the code names and the progression of our next several generations of processors." -Jen-Hsun Huang, Sept. 20, 2010 http://www.channelregister.co.uk/2010/09/21/nvidia_roadmap/

Translation: Assuming AMD doesn't misprice HD6xxx, NV has no present, so it is talking about its future instead. What else can it do, since it won't have much new stuff till 2H2011?
 

Kenmitch

Diamond Member
Oct 10, 1999
8,505
2,249
136
Nvidia quotes are amusing :)

"They (Intel) have disrupted our chipset business. The damage has been done. We've been out of the chipset business for well over a year, so if this got resolved we're not expecting to ramp back up the thousand engineers that we had working on chipsets."

Kepler's performance-per-watt advance is well underway, according to Huang. "Kepler is based on 28 nanometers. It's scheduled for production later next year. The design is progressing very rapidly. There are hundreds of engineers working on it. By the time we are done with the Kepler family, we will probably have invested a couple of billion dollars on R&D for it."
 

dust

Golden Member
Oct 13, 2008
1,328
2
71
Nvidia quotes are amusing :)

"They (Intel) have disrupted our chipset business. The damage has been done. We've been out of the chipset business for well over a year, so if this got resolved we're not expecting to ramp back up the thousand engineers that we had working on chipsets."

Kepler's performance-per-watt advance is well underway, according to Huang. "Kepler is based on 28 nanometers. It's scheduled for production later next year. The design is progressing very rapidly. There are hundreds of engineers working on it. By the time we are done with the Kepler family, we will probably have invested a couple of billion dollars on R&D for it."

You know.. a communist leader used to say similar crap : " A chicken in every plate! A Chianti for every meal!";"15000 specialists prepared to overclock the country to its full potential"
 

OCGuy

Lifer
Jul 12, 2000
27,224
36
91
You know.. a communist leader used to say similar crap : " A chicken in every plate! A Chianti for every meal!";"15000 specialists prepared to overclock the country to its full potential"

I'm sorry but I almost spit my drink on my LCD.

nV puts out a roadmap, like I have seen other companies do multiple times....and a now we are comparing them to communists?

I love this board sometimes. :)

I don't think nV's competitor is above talking about the future to distract from the present:

http://sites.amd.com/us/fusion/APU/Pages/fusion.aspx
 
Last edited:

MrK6

Diamond Member
Aug 9, 2004
4,458
4
81
Easy one, because when do they not have a refresh or higher clocked faster card.? , Both AMd and Nvidia do it.
Right, so why wouldn't that be the case this time? I said there's no reason I'd expect there not to be refreshes, and you seemed to disagree, but then answered with my own hypothesis. :confused:
I came to that conclusion because you said it yourself.
There's a reading comprehension failure. I said that performance would stagnate, I didn't say there wouldn't be any new products. For all we know, it's NVIDIA, there could be a dozen "new" products just from rebranding :whiste:.
 

Dark_Archonis

Member
Sep 19, 2010
88
1
0
Thanks for the post!

So I am guessing some sort of Fermi re-spin late 2010/Q1 2011, with a full new generation in stores H2 2011?

The fact that they are showing performance/watt on the graph means they got the hint about Fermi. ;)

I think they had "the hint" before Fermi was even released.

Well, perhaps we will see a Fermi refresh at some point. This makes it seem like 2010 is the original time line for that refresh so it either means it isn't coming(or was GF104), or that the holidays will be fun in addition to the HD6000 series.

But it seems to clearly point out that Kepler is at 28nm though. Other than the name folks have been claiming "no successor until 28nm" for some time. Nice to have a name though.

Where on earth does engadget get the time line from though? They must realize that 28nm isn't close to ready at TSMC yet. How can they claim at 28nm and "expected to go to production this year and ship in 2011" in the same breath and not be full of something fishy? They won't even get sample silicon back until 2011.

Additionally, how far along is TSMC anyway? Is the time consuming part getting up to full production capability (thus the parts at ATI/Nvidia are waiting to just push through) or is the process itself not done and thus parts can't even tape out until 2011?

Where did they get the timeline? Well if you look at the picture, the timeline was shown at a presentation. Engadget didn't just make up that timeline. Specifically, Nvidia showed that timeline at the GPU Technology Conference going on right now in Jan Jose, California.

I wouldn't hold your breathe if I was you. Maybe late 2011 if at all. It's gotta be a old PR roadmap slide. Just look at Fermi (2009) we all know what happened with that.

Clearly Nvidia had lots of problems with Fermi, and TSMC's delays didn't help either. Nothing but assumptions that Kepler would have the same problems.

Fermi did not arrive on schedule, it arrived late by Nvidia's internal timeline. Therefore, if Kepler arrives on time it will look like an early arrival compared to Fermi.

So there's nothing coming from NVIDIA for awhile then? Beyond that ambiguous allusion to "mid-life kicker" products, their roadmap is looking pretty bare.

There will be updated Fermi products before Kepler, you can count on that. That and the performance increases that Nvidia's driver team are famous for.

Since when does a lack of confirmation regarding unannounced products confirm there ISN'T new products being worked on for the current node?

Since people on tech forums began to state this :p. But yes, I agree with you.

Part of the issue with a Fermi refresh is generally you increase performance by:
1. Add more stuff. Fermi @ 40nm can't really add all that much because it's huge already.
2. Increase clocks. Fermi takes in enough power as it is, you can't increase them THAT much.
3. Go to new process, which won't be ready for awhile, so that's out of the question.

Lastly, you can do some tweaks, lower power consumption, boost clocks a bit, but it's awfully tough to get more performance out of a huge chip that takes 250watts of power as it is.

Now, hopefully they manage to pull something off, but I don't expect them to get more than 10-20% until 28nm

There is also:

4. You can have a world-class driver team like Nvidia does, and increase performance via future driver releases, considering Fermi is a radically different architecture than any previous Nvidia products. Also consider that the drivers for the AMD 5xxx cards are quite mature, and the AMD/ATI driver team is not as good as Nvidia's driver team.

Yeah, the Fermi that actually went into production is itself a refresh/redesign that went through a fair amount of revisions to get to where it is now. As you say they cant squeeze much more performance out of that design on the current process.

How many people expected the price/performance of the GTX 460, given the mediocre price/performance of the GTX 470 and GTX 480? I will answer that for you, almost nobody expected it.

The GTX 460 runs cool, and currently no AMD/ATI product can beat it in performance for its price. Not bad for a Fermi-based GPU huh? If you believe the internet gossip, all Fermi cards are supposed to be hot, slow, and expensive. I guess none of that applies to the GTX 460.

I would say let's wait and see what Nvidia brings, and let's not make so many assumptions.

They were out on time. But they were supply limited due to the issues and they had to yank out functionality in the 5000 series to get it done. Nvidia had more at stake with their chip(Tesla, Quadro lines, and gaming) and didnt have that flexibility. That said, the 40nm node issue was a fiasco for both companies.

Bingo, this is exactly what happened.

Question is though... was 40nm a fiasco or is TSMC a fiasco.. Even worse, if the sort of issues we see start to become commonplace the smaller we get both companies are going to have to have a rather large change of strategy to succeed.

The 40nm problem was neither the first issue TSMC has had, nor will it be the last. TSMC has had such issues before.

Take a deep breathe and relax :)

I doubt all the blame with Fermi's delay can be placed in TSMC's lap. I'm sure some of the delay was caused by the need to re-spin/tweak the design. One could say that AMD didn't seem to have much of a problem with TSMC's 40nm node other than supply. As far as Fermi's replacement goes it's in the old PR slide shown in the photo.

Now does nvidia have something to counter the 6xxx series launch? We'll have to wait and see I guess....I'm kinda surprised they haven't released a dual gpu card yet!

As far as the hypothetical performance of the upcoming 6xxx series from AMD it's anybodies guess....But I'd think AMD wouldn't waiste the R&D time and money if it wasn't gonna pay off for them.

Yes you could say that, but it would be incorrect.

As has already been pointed out, AMD was supply constrained to a degree and they had to remove certain features out of certain 5000 cards in order to get the cards to market.

As for AMD/ATI wasting R&D time, you'd be surprised.


When I talked to my source at AMD he said " Were gonna kick nvidia to the curb just like Apple did " so does that make you feel better about the 6xxx series hypothetical performance now :)

What I meant is that if it was a press-release as others claim then why isn't it all over the web....We all know how the Big Green Machines PR department works.

I think some of us also have a clear understanding of how the AMD/ATI PR machine works. It seems they have some PR people posting on tech forums to influence opinion.

This last portion of your post is a thread-crap and thread-derail. Please refrain from posting this manner of commentary in threads for which the topic is not about the perception of shadowy inner-workings of marketing departments.

Moderator Idontcare
 
Last edited by a moderator:

Will Robinson

Golden Member
Dec 19, 2009
1,408
0
0
Yeah but they don't get free hardware to do it I suppose,if they did....I'd be volunteering:p
Cause we all know that makes no difference in opinions here:whiste:

This post is a thread-crap and thread-derail. Please refrain from posting this manner of commentary in threads for which the topic is not about the perception of shadowy inner-workings of marketing departments.

Moderator Idontcare
 
Last edited by a moderator:

dangerman1337

Senior member
Sep 16, 2010
333
5
81
I'll post my thoughts i posted on Bit tech and just Quote it here instead of typing it all again:

If you saw Nvidia's conference yesterday you will heard of a roadmap showing two new GPUs coming in the future and another thread on Xtreme news:

http://www.tcmagazine.com/tcm/news/h...ming-next-year

http://www.xtremesystems.org/forums/...d.php?t=259492

EDIT: This site aswell: http://translate.google.com/translat...-suunnitteilla

Kelper will be on 28nm Process and Maxwell on 22nm (but will be a very big gap acording to the roadmap, maybe my system with Haswell ).

If you have been lurking around on xtremesystems news one posted called sampsa who has supposdly has sources said about the "GF110" and around the net few sites were talking about the "Fermi 2". This is what i think that the following releases will be:

GF11x/Fermi 2 = 500 series (40nm or 28nm), kelper = 600 series (28nm), Maxwell = 700 series or 800 series (22nm) if there is going to be something inbetween like a "Kepler 2" on a (28 or 22nm).

I also think that Nvidia is goint to counter AMD/ATI's follwing series GF11x/Fermi 2 will counter Northern Islands (40nm), Kelper will counter Southern islands (28nm).

Kelper from the conference has apparantly a LOT of egineers working on it which makes me think kelper is going to be a New arch like the GF100 from G200. Also GF11x/Fermi 2 could have a more "Gamer optismised" version possibly for GeForce brand and another one for HPC stuff for Quadro brand as the main problem for Fermi was that the Geforce was trying to be able to do HPC tasks aswell as gaming which lead it to be too expensive, power hungry and hot compared to Cypress due to its HPC designed shaders.

This also makes me think that both AMD/ATI and Nvidia have taken a page out of Intel and are going to start doing a "tick tock" style of devolpment and releases as die shrinks are going to get far more tougher for TSMC, GlobalFoundries to increase perfomance, Die size and Power requirements (Nvidia's CEO said that Maxwell is going to have very good Perfomance per watt by a lot in fact looks at the roadmap and if true).

By looking at this roadmap makes me think that 2011 and foward is going to be a good time for PC hardware compared to the recent dissapointing last two years (2009-2010).

So what do you all think?
 

epidemis

Senior member
Jun 6, 2007
794
0
0
Thanks for the post!

So I am guessing some sort of Fermi re-spin late 2010/Q1 2011, with a full new generation in stores H2 2011?


The fact that they are showing performance/watt on the graph means they got the hint about Fermi. ;)

The event is marketed for supercomputer users. For them performance/watt is about the only important metric.
 

Dribble

Platinum Member
Aug 9, 2005
2,076
611
136
There's almost certainly going to be a fermi refresh at some point between now and Kelper - a GTX 485 or similar - nvidia have pretty well always done this. To leave us with exactly the same high end card for best part of 2 years seems highly unlikely.

As for how good Kelper is going to be for us gamers, pretty unclear? The slide seems to be talking about double precision performance/watt which considering it's a gpu conference makes sense, however double precision is pretty well useless for gamers. It's a focus that if anything suggests it might be compromised for gamers as one of kelpers main performance targets isn't helping us.
 

Ares1214

Senior member
Sep 12, 2010
268
0
0
There's almost certainly going to be a fermi refresh at some point between now and Kelper - a GTX 485 or similar - nvidia have pretty well always done this. To leave us with exactly the same high end card for best part of 2 years seems highly unlikely.

As for how good Kelper is going to be for us gamers, pretty unclear? The slide seems to be talking about double precision performance/watt which considering it's a gpu conference makes sense, however double precision is pretty well useless for gamers. It's a focus that if anything suggests it might be compromised for gamers as one of kelpers main performance targets isn't helping us.

They were having problems with the 485. For one, most 480s are just failed 485s. When they tried to make i think it was 50,000 480s, only 6000 could even be turned into 485s, and therefore had very poor yield. Second, without radically changing things, and just adding shaders and clock speed, it will just run hotter linearly to the increase in performance. 2nd, they are taking their SWEET time just getting the barebones of their lineup out. the 6xxx series is shaping up to lap them, as they still need to roll out the GT430, and its released same time as begining of 6xxx. A fully unlocked 460 can also only reach around 480 performance, and while it may not be as hot, they would have to price it so low compared to AMD products to get any sales, in which case they would be stagnating, or possibly even losing money. They can add a "5" to the end of all their cards, but AMD is adding 1000, and thats really just saying a few more shaders here, increased clock, little tweaking cant stand up to a fully refreshed arch.
 

Keysplayr

Elite Member
Jan 16, 2003
21,209
50
91
So if Kepler is the "Tock" due out in 2011, then Fermi II must be the "Tick". The reason I think folks don't expect to see anything "so soon" from Nvidia is because Fermi was very late, making it seem like too short a period of time for the next gen/refresh release. For all intents and purposes, Fermi "should" have been here just about the time when AMD's 5870 launched in 2009. That was a year ago. Fermi has actually been here for half that length, so another architecture being released would "seem" like it's not possible or too soon. And of course, according to the logic of some, if Fermi was late, then every succeeding GPU from Nvidia must be late also. I don't think that's how things are going to work out.
 
Last edited:

Madcatatlas

Golden Member
Feb 22, 2010
1,155
0
0
They were having problems with the 485. For one, most 480s are just failed 485s. When they tried to make i think it was 50,000 480s, only 6000 could even be turned into 485s, and therefore had very poor yield. Second, without radically changing things, and just adding shaders and clock speed, it will just run hotter linearly to the increase in performance. 2nd, they are taking their SWEET time just getting the barebones of their lineup out. the 6xxx series is shaping up to lap them, as they still need to roll out the GT430, and its released same time as begining of 6xxx. A fully unlocked 460 can also only reach around 480 performance, and while it may not be as hot, they would have to price it so low compared to AMD products to get any sales, in which case they would be stagnating, or possibly even losing money. They can add a "5" to the end of all their cards, but AMD is adding 1000, and thats really just saying a few more shaders here, increased clock, little tweaking cant stand up to a fully refreshed arch.


Lets please not muddy up the waters around Nvidia gpus by giving names to cards which have never been mentioned by any reputable source.

There is no 455, no 460x2, no 475 and definetly no 485. The 480 was ment as the 512 part. All the other parts are cutdowns from this piece.

A fully unlocked 460 will not equal the current 480. That would be dreaming. It could however put up a fight against the 5870 if factory overclocked to what most good factory oc cards come as now, minimum 800mhz on the core.

If Nvidia does come with a refresh sometimes next year, id imagine a late spring release, much like the gtx480 or a summer realease like gtx460. ANd yes, they will probably then call the 512 part a gtx485. But nothing like that exists anywhere yet.
 

ginfest

Golden Member
Feb 22, 2000
1,927
3
81
Yeah but they don't get free hardware to do it I suppose,if they did....I'd be volunteering:p
Cause we all know that makes no difference in opinions here:whiste:


Why should AMD give out free hardware for folks to pimp their products and turn most of these threads into anti-NV fests? You and 4-5 others seem happy to do it for free ;)

This is a thread-crap and a member call-out. Not acceptable.

Moderator Idontcare
 
Last edited by a moderator:

Madcatatlas

Golden Member
Feb 22, 2010
1,155
0
0
Why should AMD give out free hardware for folks to pimp their products and turn most of these threads into anti-NV fests? Nvidias failures and sleazy marketing seems to result in the same, for free ;)



Fixed that for you :)

This post is a thread-crap and thread-derail. Please refrain from posting this manner of commentary in threads for which the topic is not about the perception of shadowy inner-workings of marketing departments.

Moderator Idontcare
 
Last edited by a moderator:

dangerman1337

Senior member
Sep 16, 2010
333
5
81
So if Kepler is the "Tock" due out in 2011, then Fermi II must be the "Tick". The reason I think folks don't expect to see anything "so soon" from Nvidia is because Fermi was very late, making it seem like too short a period of time for the next gen/refresh release. For all intents and purposes, Fermi "should" have been here just about the time when AMD's 5870 launched in 2009. That was a year ago. Fermi has actually been here for half that length, so another architecture being released would "seem" like it's not possible or too soon. And of course, according to the logic of some, if Fermi was late, then every succeeding GPU from Nvidia must be late also. I don't think that's how things are going to work out.

Well HD2000 > HD3000 anyone ;). That was around six months, so its not farfecthed to see a GTX580/70 that is slightly more Sqpowerful (20%-30%) and probably be slightly under 500mm Squared and less power comsumption (due to reworking the shaders abit to make them more "gamer friendly").
 

Kuzi

Senior member
Sep 16, 2007
572
0
0
So I am guessing some sort of Fermi re-spin late 2010/Q1 2011, with a full new generation in stores H2 2011?


The fact that they are showing performance/watt on the graph means they got the hint about Fermi. ;)

The Fermi we got now reached the market after many respins, and the full new generation depends on when the 28nm process will be ready. I'm thinking around Q4 2011 or so.
 

Grooveriding

Diamond Member
Dec 25, 2008
9,110
1,260
126
So if Kepler is the "Tock" due out in 2011, then Fermi II must be the "Tick". The reason I think folks don't expect to see anything "so soon" from Nvidia is because Fermi was very late, making it seem like too short a period of time for the next gen/refresh release. For all intents and purposes, Fermi "should" have been here just about the time when AMD's 5870 launched in 2009. That was a year ago. Fermi has actually been here for half that length, so another architecture being released would "seem" like it's not possible or too soon. And of course, according to the logic of some, if Fermi was late, then every succeeding GPU from Nvidia must be late also. I don't think that's how things are going to work out.

I agree with the Kepler card not being destined to be late as it is their next new 'series' card and there is no reason for Fermi's issues to infect their next architecture.

But for the here and now they have Fermi to work with. With it coming late due to whatever issues they had, amongst them clearly being heat and power consumption, this will affect whatever refresh plans they had for Fermi. I doubt we will ever see anything more of GTX 480 with a 512 shader part.

They may release a fully unlocked GF104, but that won't even be as fast as the 480 is now, they may release a dual gpu card, but that won't be as fast as whatever AMD is going to release as the 6990/6970 or whatever they end up calling their new dual-gpu card.

This announcement confirmed that there is no new architecture or series coming from nvidia until September or later next year. So that leaves us with refreshes of Fermi. I'm sure the refreshes will offer something more than the performance NV has now, I would be surprised to see a single GPU that is faster than a GTX 480 though, probably going to be a multi-gpu card.

I've never seen a refresh of a current series compete with a brand new series. It would be akin to the GTX 285 being faster than a 5870, it's not going to happen, although the GTX 295 was faster than a 5870 at launch. If they drop a multi-gpu card, it will likely be faster than a 6870, and will probably be a good buy because they'll have to price it fairly low to stay under wherever AMD is pricing their new dual-gpu flagship and not too far from wherever AMD's single-gpu flagship is selling at.

I think this will be their problem in the present, they can't offer an exciting card using refreshes against AMD who is going to be selling a card on a new architecture. They'll probably gain traction from selling cards on their cheap price points, not dominant performance.

In short, yeah they'll definitely release new cards between now and 2H 2011, but they're not going to be the sort of cards you can get excited about and look forward to seeing benches of. I've seen a ton of refresh parts, I know what to expect, 5 more fps.
 
Last edited:

Creig

Diamond Member
Oct 9, 1999
5,170
13
81
Is it just me, or does being able to accurately plot the expected performance of a chip two full generations away from production on a node that doesn't yet exist seem a bit... optimistic?
 

Kuzi

Senior member
Sep 16, 2007
572
0
0
As per original post:

"Must be going by the "if not delayed by TSMC's 40nm problems"

I believe it was a combination of TSMC's troubled process and Fermi being a nightmare to manufacture with reasonable yields. As you know ATI released a 40nm card about a year earlier than Fermi, the Radeon 4770.