-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Jen Hsun on Kepler (CES 2012)
- Thread starter Havana_UK
- Start date
This may be the GTX 580 replacement which will probably show up as a 760. 576 cores@950Mhz doing estimated ~2 GFLOPs by the looks of it.
If the specs are correct then this might match the 7970.
Soldato
- Joined
- 21 Sep 2009
- Posts
- 5,267
- Location
- London
Nvidias dilly dallying will work in favour for AMD but not for prices for end users. Is it possible they've seen AMDs 7970 and had to go back to the drawing board? Just like AMD did with Bulldozer after Intel released Sandybridge?
Caporegime
- Joined
- 9 Nov 2009
- Posts
- 25,773
- Location
- Planet Earth
Kyle Bennett from HardOCP has said closer to June from what he heard from his sources.
Possible, but I think it's unlikely. NVIDIA probably aimed much higher than their own 580s for the top end parts. Which would give them a comfortable lead. I doubt they 'have planned a top end GPU which is not more than 30% faster than their existing 40nm 580.
If the leaked specs for the GK104 are true then
905 * 576 * 4/ (1540 * 512 * 2) = 1.32224
So it would be about 32% faster than the 580.
If the leaked specs for the GK104 are true then
905 * 576 * 4/ (1540 * 512 * 2) = 1.32224
So it would be about 32% faster than the 580.
There's no way that nVidia could go "back to thr drawing board" and produce anything significantly different in the time frames we're talking about. Developments like this are 18 months+ I would think.
Tweaks, maybe. Back-to-the-drawing-board changes, no chance.
I imagine they're having difficulties with the 28nm process, but the architecture is fixed. This would also support the release of a mid range (smaller silicon) device sooner than the top end.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
The 40nm 4770 to test the waters was a 4770, it came after the entire HD4000 was out, 4850/70 first, all the way down to the low end, then came a midrange 40nm version.
We're not really talking about process node here, but architecture, usually when a company launches an architecture the big chip comes first. Its the most problematic and hardest to get working and generally speaking its the architecture, teams work on the other parts concurrently but not till the architecture is "sorted", then you work on the individual chips but exclusively the high end new architecture has come first for some time.
Nvidia didn't launch the 480gtx, they launched "fermi" and the first card was the 480gtx, the first Fermi + 1 was the 580gtx, the first Evergreen was the 5870, the first HD4000 was actually IIRC the 4850, then the 4870, but same core(and I believe delayed mostly for gddr5 availability, and I think it was weeks, it was a while ago). AMD have done that before, x800pro availability before the x800xt and xt pe, but its always the high end chip first.
Hence when someone refers to "Kepler" being available its relatively natural to take that to mean the high end, this time however that is incredibly unlikely.
That simply isn't how gpu's work, AMD didn't rework Bulldozer after Sandybridge launched and neither can Nvidia rework what they are doing.
Bulldozer taped out BEFORE Sandybridge launched, with cpu's its usually in the range of 12-18 months from tape out to launch. With Gpu's its no where near as long, but still pretty long, The 7970 seemingly taped out around March-ish, though I can't recall exactly when, was likely ready to go around september but was waiting on TSMC an extra few months. If Nvidia launches in call it June, the chip will have taped out before the 7970 launched, and any "significant" change to the chip would delay tape out by months on end. You can't just up the efficiency on a whim, nor magically make it capable of higher speeds by a minor tweak you can do in a few weeks.
Well if memory serves the low end Kepler was coming out then, with the high end being reserved until at least Q3 2012.
All that was coming out first were laptop and onboard solutions with the high end coming Q3.
That's what happens when you're winning I guess, you get to become tardy.
They're going to need to wake up though. Three months of lost sales is too much, let alone 8 !
nVidia aren't going to be losing a whole lot based on not having a high end part to compete with AMD for a few months. It will account for such a small proportion of their sales. They sell far more into OEMs like Dell than they do to you and I, that's where the real money is.
Also look at what nVidia are doing at the moment, they've got a lot of focus on the mobile market with Tegra and other technologies. Again, volumes there are potentially substantial and dwarf the opportunities in high-end CPUs for our market, and arguably even the PC market as a whole.
Also look at what nVidia are doing at the moment, they've got a lot of focus on the mobile market with Tegra and other technologies. Again, volumes there are potentially substantial and dwarf the opportunities in high-end CPUs for our market, and arguably even the PC market as a whole.
In al seriousness, would mid-range Kepler be a BAD thing?
Ok - new high end GPU parts are damn exciting and we all love huge numbers, but they are not where a large volume of sales could reside. Very few of us need 7970 performance now on most resolutions. What if nVidia release a mid-range card at or around 580 performance with headroom for overclocking and 2+GB of VRAM at a cracking price-point?
I think it would sell like hot cakes and if they can get it out in Q1 or early Q2 I would certainly be interested.
Ok - new high end GPU parts are damn exciting and we all love huge numbers, but they are not where a large volume of sales could reside. Very few of us need 7970 performance now on most resolutions. What if nVidia release a mid-range card at or around 580 performance with headroom for overclocking and 2+GB of VRAM at a cracking price-point?
I think it would sell like hot cakes and if they can get it out in Q1 or early Q2 I would certainly be interested.
Kepler has more [mobile] design wins than Fermi: http://www.fudzilla.com/component/k2/item/25508-kepler-has-more-design-wins-than-fermi
More evidence that nVidia are pushing hard in the mobile space, hinting at improved power efficiency in the Kepler architecture.
More evidence that nVidia are pushing hard in the mobile space, hinting at improved power efficiency in the Kepler architecture.
Saying they would show it off and then deciding not to is not good news. We can think whatever we want about the state of Kepler right now but they are the ones who really know what's happening, so to say its ready to show and then not show it is only good news for AMD.
It's either not good enough to show, too rushed and buggy to show or there was nothing to show and it was just taking a bit of attention away from AMD.
Will Nvidia fall a generation behind in the next couple of years or bow out of desktop race first?
It's either not good enough to show, too rushed and buggy to show or there was nothing to show and it was just taking a bit of attention away from AMD.
Will Nvidia fall a generation behind in the next couple of years or bow out of desktop race first?
Soldato
- Joined
- 10 Aug 2006
- Posts
- 5,207
I am anxious to see what Nvidia pulls out the bag. I am hoping we have a good bit of rivalry this year between AMD and Nvidia. I don't like to see one card in particular dominate the scene - I think it's healthier for the consumers to have more a of a choice and competitive pricing scheme to match, especially at the top end.
Soldato
- Joined
- 30 Nov 2011
- Posts
- 11,527
thing is, GTX 580 (et al) is obviously still selling quite well, so why kill the cash cow when you don't really need to
either that or GTX780 will be £650 when it comes out (it will almost certainly be $650-$700 stateside and what we keep seeing is that $ = £ by the time they get here)
either that or GTX780 will be £650 when it comes out (it will almost certainly be $650-$700 stateside and what we keep seeing is that $ = £ by the time they get here)
In addition, while the 7970 IS the faster of the two, it's not THAT much faster in real world terms.
7970's biggest advantage right now is its apparently massive headroom for overclockability (is that even a real word?).
From purely a response on forums like this perspective Nvidia couldn't really win, what ever they did they would get slammed for. we all remember the wood screws incident, I would have loved to have been a fly on the wall for the conversation with who ever decided that was a good idea. I think a simple "be patient" was probably the sensible option this time round, even though we all want to know when Kepler is coming.
As for performance I reckon Nvidia have a bit of an issue, I reckon its fast but just as AMD seem to have found, power usage is high (which i think is due to the process not so much the design) so what to do launch a 760 with superb performance but leave no room TDP wise for the rest of the lineup, or downclock it so it fits into the power usage envelope and leave room for the chips to come in above it. i do feel that this is what AMD have done with the 7970, it clearly could have been faster (higher clocked) but then where would the power usage be, and where do you then go with the twin GPU part.
Nvidia would have an even more difficult time seeing as the 760 would have two or even three/four (if you count the twin gpu parts) above it, if it is already getting close to power usage limits then where do you go.
anyway all this is just conjecture with nothing to back it up, but of course one mans opinion is just a valid as another mans.
unless he disagrees with me course
As for performance I reckon Nvidia have a bit of an issue, I reckon its fast but just as AMD seem to have found, power usage is high (which i think is due to the process not so much the design) so what to do launch a 760 with superb performance but leave no room TDP wise for the rest of the lineup, or downclock it so it fits into the power usage envelope and leave room for the chips to come in above it. i do feel that this is what AMD have done with the 7970, it clearly could have been faster (higher clocked) but then where would the power usage be, and where do you then go with the twin GPU part.
Nvidia would have an even more difficult time seeing as the 760 would have two or even three/four (if you count the twin gpu parts) above it, if it is already getting close to power usage limits then where do you go.
anyway all this is just conjecture with nothing to back it up, but of course one mans opinion is just a valid as another mans.
unless he disagrees with me course
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
Well that is pretty much the entire point, its not the midrange most likely but the low end, and that might get delayed by a couple months as well. IE if baby Kepler was due March or so, and needed a respin, that would push it to late May or.... June and wouldn't be unsurprising.
That doesn't mean the GK104 will be pushed back as well but, this is generally why you work primarily on one design, getting it working on the process, finding out how to tweak that core for the process then the teams that have laid out and done what they can for the lower end parts also add these tweaks(where appropriate) to get working well on the process.
Its unlikely that if the GK107 had an issue that required a respin, that the GK104 somehow wouldn't also have that issue.
As for why the midrange isn't bad, ultimately its not and its good for users to have the same performance at a lower price point. However there are reasons no one is that "interested" in it, the same way the 5770 was a nice card but it wasn't exciting, ultimately you can have bought roughly 580gtx performance for two years, from overclocked 480gtx's through to current 580gtx's, and in a while midrange Kepler's...... new card, lower price point maybe, but a performance level that was pretty much achievable 2 years ago, and by the time it launches, maybe closer to 2.5 years. Its not bad its just, not new, its not particularly interesting.
Also with EOL pricing on the 480gtx, you could have picked up that kind of performance for under £200 ages ago now, and the 560 is unlikely to be under £200 anyway.
It's not anti Nvidia, AMD and Nvidia midrange and low end parts are essentially a performance level you've seen before at a new price point, and not always brilliant prices these days.
Think about it a 4850 was £125 on launch, a 5770 was almost the same price and only a bit faster because of the increase in core sizes on the hd5000 and increasing process problems.
The gen before that going from 3870 to a 4670 which was WAY cheaper was marginally interesting but still a bit...... meh.
either that or GTX780 will be £650 when it comes out (it will almost certainly be $650-$700 stateside and what we keep seeing is that $ = £ by the time they get here)
650-700 its impossible
600 maxs as amd will have new series already
650-700 its impossible
600 maxs as amd will have new series already
I'm hering May as the most likely date.
Would not be unknown for NVIDIA to pull it out the bag though and come as early as March, but May looks likely.
Problem is if NVIDIA don't get it out soon AMD could counter them with 7990 to compete with Keplar and move price down on 7970 possibly.
With 7950 3GB due next month and 7950 1.5GB due in March ATI will have NVIDIA completely owned as it would not make any sense to buy a GTX 570 and GTX 580 with faster cards from ATI for same money.
69xx series already has 560 range covered, though those will be gone around March and 78xx will come out, the 78xx I am not quite sure of yet, its looking like it may be around 69xx performance or marginally slower but with far less power usage and similar price.
Would not be unknown for NVIDIA to pull it out the bag though and come as early as March, but May looks likely.
Problem is if NVIDIA don't get it out soon AMD could counter them with 7990 to compete with Keplar and move price down on 7970 possibly.
With 7950 3GB due next month and 7950 1.5GB due in March ATI will have NVIDIA completely owned as it would not make any sense to buy a GTX 570 and GTX 580 with faster cards from ATI for same money.
69xx series already has 560 range covered, though those will be gone around March and 78xx will come out, the 78xx I am not quite sure of yet, its looking like it may be around 69xx performance or marginally slower but with far less power usage and similar price.