Closed

What is Nvidia's counter to the HD 6xxx series GPU's?

What is Nvidia's counter to ATI's Southern and Northern Island HD 6xxx series graphics cards? Is a GTX 5xx series in the works? Fermi was sort of a flop, the GTX 480's are known to reach 100*C in cases with poor airflow, consume more than 300W of power, and was barely faster than the HD 5870. Although it does handle tesselation 75% better than the ATI cards.

I wonder if the GTX 580 will beat out the HD 6870 by more than a fair deal.
332 answers Last reply
More about what nvidia counter 6xxx series
  1. No one knows.

    But don't be surprised if they simply rebrand some old cards and call them new.
  2. I stand corrected.
  3. Quote:
    www.guru3d.com/news/nvidia-kepler-is-succesor-to-fermi--due-2011-already/
    That is nvidia's counter.


    Will Kepler and Maxwell max out Crysis? What about Four flagship Maxwell chips in SLI? Will that finally conquer Crysis?
  4. The cards that goes against each other at that time will be much much better in perf.
    We have a whole shrink coming, instead of just half a node, plus the introduction of HKMG, which helps with power/cooling/thermals
    Add this all up, and that gen will be very powerful indeed, from both camps
  5. Another one of these threads. To be honest, Nvidia doesnt have much to answer against 6xxx. Kepler will arrive later in 2011, along with Southern Islands, 7xxx. Besides a fully unlocked GF104, dual GF106 part, and maybe some rebranding and tweaking, Nvidia doesnt have much from what i see, atleast for now. You do have to realize, Fermi went pretty bad, and came late 6 months, and it still coming! A year after the release of 5xxx and Fermi is still coming, so its undisputable they are behind on time right now.
  6. Take into consideration, that they could add more to Kepler, up the clocks, and still have decent thermals/power usage.
    Expect the same from ATI
  7. JAYDEEJOHN said:
    Take into consideration, that they could add more to Kepler, up the clocks, and still have decent thermals/power usage.
    Expect the same from ATI


    Well its a die shrink, id expect around 75% performance increase, atleast. But even Kepler didnt seem terribly impressive on that chart...
  8. ares1214 said:
    Well its a die shrink, id expect around 75% performance increase, atleast. But even Kepler didnt seem terribly impressive on that chart...

    40nm is currently considered a half node, 45nm and 32nm being full nodes.
    At 28nm, thats going to the next half node, not stopping at 32nm, or the full node, so its more than weve seen lately, plus the HKMG on top of that
  9. Quote:
    ^And what special info you saw in that chart which makes you conclude Kepler is unimpressive.


    In the chart it shows Kepler with 3x the "CUDA Performance Per Watt" than fermi. Well for one, the fact they added in "Per Watt" should tell you something. They just wanted to make fermi look bad, and this look good, which is likely. If it beat it by 3x the performance, they wouldnt of had to add the watt part in. Secondly, its CUDA performance. Something that they can manipulate the most, and something that cant be compared to AMD. CUDA performance has nothing to do with real performance, so once again, i doubt they would put that if they had 3x, or even 2x the performance. Lastly, it shows Fermi in 2009. No. Fermi came out 2010, mid 2010 at that. If they have that on the chart, then they must be refering to another date, which pushes Kepler back a decent bit. So even if it is "3x the CUDA Performance Per Watt", which its a Nvidia chart, all company charts showing performance are exagerated, AMD could show the X4 955 beating the i7 930 if they wanted; its not that impressive. 3x the performance per watt, thats getting slightly better, but the fact they have to add per watt in scares me, as it might just be to make fermi look worse.
  10. OK, lets go with the charts
    In 2009, Fermi obviously wasnt ready, so doing 4x the perf of something nVidia wouldnt even release means what?
    As far as cuda goes, thats SW, not HW, and SW always gets better without changes to HW alone, besides once again, the starting point for Fermi, which is dubious at best, since it was released months later.
    And since this is all speculation and maybe just simulation if that at this point, I still wonder why nVidia is making these claims
    If they miss these marks, itll make them look very bad indeed, especially with Fermi coming in so late, so hot, needing alot of power, and not hitting its clocks/perf
  11. Quote:
    ^Stop bashing fermi.Each fermi gpu smokes Ati evergreen gpu at there respective price point.And if you say they run hot,breaking news for you till date i am yet to read a thread where a user has complained that his fermi gpu is overheating or burned.


    It doesnt really matter how you look at it, Fermi was a pretty big failiure until the GTX 460 came along, and even that wasnt enough. Fermi was released 6 months late, and its still being released, 12 months after. It was priced too high, ran too hot, and used too much energy, and didnt even give that big a of performance increase. Take the GTX 480 for example. Beats the 5870 by around 15%. Comparing the cheapest 480, and the cheapest 5870, (not including MIR), the cheapest 480 is $450, the cheapest 5870 is $350. So it costs 30% more, and gives 15% more performance, runs 10-15 degrees hotter, and uses over 100 watts more at load. That sound like a good deal? Not to me it doesnt! The 470 was a bad deal, but when it dropped to $300, it gets about the same price/perf as a $350 5870. So your telling me, fermi wasnt a failiure? I think even Nvidia admitted Fermi was a failiure.
  12. JAYDEEJOHN said:
    40nm is currently considered a half node, 45nm and 32nm being full nodes.
    At 28nm, thats going to the next half node, not stopping at 32nm, or the full node, so its more than weve seen lately, plus the HKMG on top of that


    When will the laws of physics make Moore's law no longer feasible? How much smaller can a transistor get before it can't get any smaller?

    What will hardware designers do to improve performance? 40nm then 32, now 28? I don't think we'll ever see picometer chips. The laws of physics forbid it.
  13. Currently, theyre toying with 11nm as the cutoff point
    It may come down to using a different material after this, as silicon may become too expensive to work with, or even impossible beyond this point
    Theories are 7-10 years, and we will need something different
  14. JAYDEEJOHN said:
    Currently, theyre toying with 11nm as the cutoff point
    It may come down to using a different material after this, as silicon may become too expensive to work with, or even impossible beyond this point
    Theories are 7-10 years, and we will need something different


    Graphene...signle digit nanometer, 1 THz theoretical... :ange: :sol:
  15. Maybe GTX 485 or GTX 495.
  16. Mudit Sathe said:
    Maybe GTX 485 or GTX 495.


    I doubt a 495 will be possible for some time. Fermi-100 just doesnt fit to dual GPU, so it should be a while before we see anything dual. Maybe dual 450 or something like that. 485 is very likely, although they have gotten very low yields with it, i believe, ill check, but i believe 8000 of every 50,000 480's were able to turn into 485's, and for a performance boost that would be linear to heat and power, i just dont really think its practical. The 480 is already too expensive, and im skeptical a 485 would beat a 6970, so without some serious cutting down, i dont think it would compete very well. Nvidia's best bet is to do as much with GF104 as possible. Fermi didnt seem to scale linearly. GF100 was powerful, GF104 was powerful and efficient, and then we have GF106 losing a lot of that, and id assume GF108 as well. The more they took off, it got worse than i thought it should have gotten. The sweetspot was GF104, so they should expand it as much as possible. That would be a nice die shrink, and the GTX 460 is much more efficient than the 480 in almost every way. As far as die size, not quite to 5870 efficiency, but if they make an effort to go smaller, i think they would be far more sucessful.
  17. If nVidia has their tweaks done right, and the maturity of the 40nm process is good enough, they could go full Fermi with somewhat decent clocks
  18. JAYDEEJOHN said:
    If nVidia has their tweaks done right, and the maturity of the 40nm process is good enough, they could go full Fermi with somewhat decent clocks


    Well, they cant really tweak that much, and yeah, they can clock it higher, but still, if you see what they were unlocking, it would take some fairly serious clock increases to face a 20% deficit the 6970 seems to put on it. We'll see about the 6970, but Nvidia will likely have the same problems if they keep going this big.
  19. If we look at the past, its easy to see they too will need a 2 chip card once again, tho, theyve once again lost in the perf/density to their last gen, tho cleaning up with the 460 has helped in this regard
    So, a fairly cleaned out full Fermi with decent clocks yes, a duo? Not so much
  20. From your link at HWC
    "All of this innovation is exciting news for enthusiasts, but one has to wonder whether or not either of the dual-GPU boards will be available for purchase any time soon. Since neither card was shown with a cooling solution, plenty of room for speculation has been left and this news poster wouldn’t be surprised that if they are sold at retail, they’d be packaged with a water block. "

    No cooling solutions? It was June, nothing since?
    This to me looks more like certain cpus used for ocing only, not for typical usage
  21. JAYDEEJOHN said:
    From your link at HWC
    "All of this innovation is exciting news for enthusiasts, but one has to wonder whether or not either of the dual-GPU boards will be available for purchase any time soon. Since neither card was shown with a cooling solution, plenty of room for speculation has been left and this news poster wouldn’t be surprised that if they are sold at retail, they’d be packaged with a water block. "

    No cooling solutions? It was June, nothing since?
    This to me looks more like certain cpus used for ocing only, not for typical usage

    There was also some discrepancy as to what GPU's those actually were as well, 465's, 470's or even 480's and if it were to come to production you would have thought that would have been something said or seen by now.
  22. Mousemonkey said:
    There was also some discrepancy as to what GPU's those actually were as well, 465's, 470's or even 480's and if it were to come to production you would have thought that would have been something said or seen by now.


    From the description, it appears to be 2 GTX 470s together.


    if you look at the card, you can read "2x GTX 470 Prototype".
  23. Just some things to consider
    First, costs
    The density perf war is won by ATI, no contest
    Second, power, again, won by ATI, and presents problems for both companies going dual
    Going smaller has shown, unless alot if cut out of Fermi, its lil bros lose in scaling downwards in perf, unlike ATI, which shows great scaling going downwards/smaller, and even equal perf in DX11

    Last but not least is cooling, where once again, ATI wins, and nVidia will surely be battling a cool, quiet scenario as compared to the red camp, which can determine just how much perf they can achieve, not whether they can do it at all
  24. Nvidia is letting ATI make the $ 1000, 1100, and 1200 dollar video cards right now, did you notice the amd fanboys who whine about "1000 intel cpu's" don't have anything bad to say about these gold plated monstrosities ?
  25. ares1214 said:
    From the description, it appears to be 2 GTX 470s together.
    http://www.hardwarecanucks.com/wp-content/uploads/dual11.jpg

    if you look at the card, you can read "2x GTX 470 Prototype".

    I know what the card says because I posted this way back in June when the story first came out but if you look at the PCB there almost seems to be spaces for another RAM module for each GPU and there was some other pics floating around where the GPU's were not marked up as GF100's.
  26. notty22 said:
    Nvidia is letting ATI make the $ 1000, 1100, and 1200 dollar video cards right now, did you notice the amd fanboys who whine about "1000 intel cpu's" don't have anything bad to say about these gold plated monstrosities ?

    If the rumours of ATi's rebranding/renaming are true it's going to be interesting to see how well that goes down with the faithful as well. ;)
  27. I think price n perf, all will be happy
  28. notty22 said:
    Nvidia is letting ATI make the $ 1000, 1100, and 1200 dollar video cards right now, did you notice the amd fanboys who whine about "1000 intel cpu's" don't have anything bad to say about these gold plated monstrosities ?


    Aside from the Ares, or other manufacturer cards, which isnt even really a AMD video card, its Asus/XFX, where do you see $1000 video cards? The highest i see is the 5970 which is $599. Its like me saying the 480 is selling for $650 because one manufacturer decided to put the waterblock on. No stock AMD card is selling close to $1000. And i doubt Nvidia is "letting" AMD do anything.
  29. ares1214 said:
    Aside from the Ares, or other manufacturer cards, which isnt even really a AMD video card, its Asus/XFX, where do you see $1000 video cards? The highest i see is the 5970 which is $599. Its like me saying the 480 is selling for $650 because one manufacturer decided to put the waterblock on. No stock AMD card is selling close to $1000. And i doubt Nvidia is "letting" AMD do anything.


    http://www.google.com/products?q=Sapphire+5970+4GB&hl=en&aq=f
    http://www.google.com/products?q=asus+5970+ares&hl=en&aq=f
    http://www.google.com/products?q=xfx+5970+4gb&hl=en&aq=f
    This thread should be titled will AMD be able to answer Nvidia. What are the rumors again, oh they are bringing some form of 3d, better dx11 performance, and possibly a card to compete with in the 200 dollar range ?
    There is already a next generation card that does all that = The gtx 460 :)
  30. ares1214 said:
    Aside from the Ares, or other manufacturer cards, which isnt even really a AMD video card, its Asus/XFX, where do you see $1000 video cards? The highest i see is the 5970 which is $599. Its like me saying the 480 is selling for $650 because one manufacturer decided to put the waterblock on. No stock AMD card is selling close to $1000. And i doubt Nvidia is "letting" AMD do anything.

    But when the Asus Mars card was the most expensive thing out there the little fact that it was not a stock Nvidia card didn't stop some from saying that Nvidia had the most expensive card out there, I don't think you were around then though so you might have missed those fun and games.
  31. notty22 said:
    http://www.google.com/products?q=Sapphire+5970+4GB&hl=en&aq=f
    http://www.google.com/products?q=asus+5970+ares&hl=en&aq=f
    http://www.google.com/products?q=xfx+5970+4gb&hl=en&aq=f
    This thread should be titled will AMD be able to answer Nvidia. What are the rumors again, oh they are bringing some form of 3d, better dx11 performance, and possibly a card to compete with in the 200 dollar range ?
    There is already a next generation card that does all that = The gtx 460 :)


    Those are all factory altered cards with 4 gigs, so no. Normal stock AMD cards max at $714.99 at newegg. And wow, so we should have just stopped at the model T? I mean, we already had a car, why improve on it? Thats exactly what your saying. Why should AMD improve on Nvidia, Nvidia already has a card out, and thats BS, we would move no where. Im getting pretty tired of hearing NV fanboys saying AMD is behind and theres already a card that does everything AMD is trying to do with Barts, and that the 460 can OC to 5870 levels. Its getting pretty old, and we all know how much BS it is.
  32. Well, if its sellable, then its made and sold, simple as that.
    As for nVidia letting ATI do anything is proposing nVidia controls the entire market, which since the 4xxx series, nVidia has been extremely generous
    As to pricing, these cards will show their worth, especially against the 460, unfortunately, the only solid price perf Fermi card
    Maybe ATI was being generous by leaving a hole in their lineup ;)
  33. Mousemonkey said:
    But when the Asus Mars card was the most expensive thing out there the little fact that it was not a stock Nvidia card didn't stop some from saying that Nvidia had the most expensive card out there, I don't think you were around then though so you might have missed those fun and games.


    I wasnt here, but i was "around". At that time, i think i was coasting along with a 8800. Then i got a 4890 eventually. So i cant really speak on that, but ive always considered the 295 a pretty bad deal anyway, however i dont think just because Asus makes a card and sells it for XXXX amount means Nvidia is selling it for XXXX, considering they dont have much control over it.
  34. You mean like AMD fanboys spotting a Kepler slide and analyzing the context of a sentence. Or spreading fud with made up manufacturing yields ? Spreading rumors as facts ? yeah its sickening.
  35. JAYDEEJOHN said:
    Well, if its sellable, then its made and sold, simple as that.
    As for nVidia letting ATI do anything is proposing nVidia controls the entire market, which since the 4xxx series, nVidia has been extremely generous
    As to pricing, these cards will show their worth, especially against the 460, unfortunately, the only solid price perf Fermi card
    Maybe ATI was being generous by leaving a hole in their lineup ;)


    Yeah, last time i checked, AMD had the market share lead, so NV isnt letting AMD do anything. Intel might be letting AMD do things, but NV isnt. NV is losing as far as im concerned, but i guess that just makes me another "AMD Fanboy". :pfff:
  36. notty22 said:
    You mean like AMD fanboys spotting a Kepler slide and analyzing the context of a sentence. Or spreading fud with made up manufacturing yields ? Spreading rumors as facts ? yeah its sickening.


    Your saying i shouldnt analyze it? I should just read graphs, and not know what its talking about? That should get me far in life...And the manufacturing yield ill try to find the article. Spreading rumors as facts? Or speculating? Guessing? No where have i said something that isnt pretty certain is 100% true, in fact i generally leave disclaimers. I think the fact your completely missing the point is more sickening...
  37. Goodness, people seems to forget that Kepler is limited by the 28nm process more than anything.

    Considering how much trouble TSMC went though for 40nm, and how risk-production for 28nm is only starting in Q4, the earliest time Kepler can come out is ~April when everythings goes smoothly. Unless its a simple die-shrink, I doubt it will come before June, In which AMD should be more likely than not be ready for HD7000.
  38. You analyzation and 'impressions' were based on the CEO using Cuda performance per watt, which is a term being thrown around constantly lately. Its directly related to future computational performance/cost. You might as well said you read body language.
    You basically stated you saw the announcement slide = Ares not being impressed.
  39. Its a fact that nVidias been extremely generous with the halo product, since theyve held it for a few short months over the last few years

    Its a fact nVidia is boasting a card 1 year out already

    Its a fact Jenson acts as if the in between tweaks are mere nothings, easily done, as if we will see easily affordable halo on down tweaks coming soon

    Its a fact using more silicon, spending more on R&D for huge chips with lower perf/mm is more costly

    Its a fact doing this means lower yields, higher costs, and sometimes late to market solutions
    Other than that, nVidia is doing a good job, in consideration of all its hurdles/promises thus far, but I question to what end?
  40. notty22 said:
    You analyzation and 'impressions' were based on the CEO using Cuda performance per watt, which is a term being thrown around constantly lately. Its directly related to future computational performance/cost. You might as well said you read body language.
    You basically stated you saw the announcement slide = Ares not being impressed.


    If I saw a chart with AMD showing Perf/watt, the 5870 would beat the 480, and by quite some bit. I wouldnt be impressed. Same with this. For one, if it that far out, how do they really know? Are they just guessing? Ive never really seen a company guess "low" for something like this. Such as Barcelona. If they are guessing, they have no idea, if they arent guessing, then its probably close to the truth. Although if it was so much more powerful, would they really need to add the "per watt" part in? The 5870 is more powerful per watt, but it doesnt mean its more powerful. Also its CUDA performance, something AMD doesnt have, they can manipulate it the most, and is more software than hardware. So based off that, how do you know they didnt just cut power down 50%, make it 100% more powerful, and make CUDA software 50% more efficient? We are already above 3x more powerful. The jump from 55nm to 40nm yielded 55-70% performance gains from AMD(4870-5870), and 50-70% performance gains for nvidia. And thats cutting down the process 38%. From 40 to 28 is 43%, so id say its very reasonable to expect ATLEAST 75% performance increase from both AMD in 7xxx and Nvidia in 5xx. So overall, im not saying its extremely unimpressive, id need more info, but i dont find it as impressive as most people who just see "3x...".
  41. Amd is promising 10% more performance with 10% less power.
    No pressure.
  42. Ummmm....
    Lets not forget the addition of HKMG, which also brings its own power/perf additions
    If some tweaks are dont on both sides, HW alone, I could see a 2x perf increase @ 28nm w/HKMG
    That being said, once under their (nVidias) TDP wishlist, you usually will see a better perf/watt increase, often allowed by a shrink, besides what youd normally see
    In other words, Fermi/Kepler will benefit hugely from this die shrink, and if we see changes in ATIs new 6xxx series, we may see some astounding density/perf/mm numbers @ 28nm, tho where the tailing off of scaling to me is an unknown
  43. notty22 said:
    Amd is promising 10% more performance with 10% less power.
    No pressure.






    Well, since this is all speculation, adding over ten% shaders/plus tweaks, if you go by the 4=5 - 1.5% perf, we also dont know the cost/watt of this change, if true.
    That makes a 10% perf increase using 10% less power within scope

    The difference here tho is, its already being done on silicon, on a mature process, whereas nVidias proposals are reaching into the future on a new process, with new doping, being run on simulators only, provided the SW gels with the HW as well
  44. Due to the PCIe 2.0 specifications a card above 300 watts is out of the question. NVIDIA is a member of the PCI Express SIG so they can't make a 470x2 unless it's seriously crippled/underclocked. The 480 already went a little too far.
  45. mosox said:
    Due to the PCIe 2.0 specifications a card above 300 watts is out of the question. NVIDIA is a member of the PCI Express SIG so they can't make a 470x2 unless it's seriously crippled/underclocked. The 480 already went a little too far.

    But it could totally be an AIB product "endorsed" by Nvidia, think Ares, except more wide-market.
  46. notty22 said:
    Amd is promising 10% more performance with 10% less power.
    No pressure.








    Not funny. Doesnt make any sense either.
  47. ares1214 said:
    Not funny. Doesnt make any sense either.

    Neither does ANY of your math.
  48. notty22 said:
    Neither does ANY of your math.


    And how does my math not make sense? Did you pass math class? If thats a yes, then it should.
  49. Timop said:
    But it could totally be an AIB product "endorsed" by Nvidia, think Ares, except more wide-market.


    Maybe not, apparently Nvidia weren't too happy about the Mars card which is why it never gets listed in the driver .ini file if you go looking.
Ask a new question

Read More

Graphics Cards Gtx HD Nvidia Graphics