GTX 660M = GK107 (according to driver .ini and SKU-string spekulations), should be 20% faster than GTX 560M (GF116) according to Charlie
GTX 675M could be a rebranded GF114: http://forum.beyond3d.com/showpost.php?p=1604582&postcount=8305
GTX 675M could be a rebranded GF114: http://forum.beyond3d.com/showpost.php?p=1604582&postcount=8305
I can't remember where I heard it, but I believe someone from AMD mentioned that there would be no mobile Tahiti, so yes, most likely.
Ohhh.... Trinibwoy is taking the thread off topic.
The backseat moderator is gonna get you! (and me too)
Yes. Pitcairn and Cape Verde, with different code names, are the GCN mobile with the bottom being matured 40nm designs, hopefully with some updates.
http://semiaccurate.com/2011/08/24/amds-28nm-mobile-lineup-leaked-too/
Wimbledon should perform pretty close to a desktop 6870.
The backseat moderator is gonna get you! (and me too)
Yes. Pitcairn and Cape Verde, with different code names, are the GCN mobile with the bottom being matured 40nm designs, hopefully with some updates.
http://semiaccurate.com/2011/08/24/amds-28nm-mobile-lineup-leaked-too/
Wimbledon should perform pretty close to a desktop 6870.
DegustatoR
Veteran
This is an example of a random statement, pulled out of thin air. You can't say anything about how solid a new architecture is until you've compared it to it's direct competitors.
Eh? You most definitely can. Whether an architecture is solid or not is really about how it handles various real world work loads, much more so than it's relative standing to the competition; an architecture can be solid and still not be competitive. The technology has a *purpose* and as long as you understand the underlying architecture well enough you can make statements like that in absence of any direct comparisons; there are absolute performance results, and there is direct architectural analysis, this isn't just about a horse-race between AMD and nVidia; it's about the technology and how it works and what work loads it needs to cope with and how well it does that. Relative comparisons are only useful when they increase our understanding of the underlying chips. This is the architecture forum and not the industry forum right?
Sorry for continuing the off-topicness. Back to lurking.
This was posted on HardOCP - http://hardforum.com/showpost.php?p=1038306860&postcount=1
Could be a massive practical joke.
You would think Charlie is the least gullible person in the world to fall for any of nVidia's tricks. Has to be a doozy to get by his skepticism (and contempt). He said he's working on a big article to explain the "nuances" of the situation and why "nVidia will make wall st look like ghandi". Should be a classic.....
Eh...classic isn't the word that springs to mind. Hope it has mucho nonsensical techno-babble though, we totally need a new generation of misconceived notions about these chips, it's pretty clear that the evil die-size may be on its way out. Oh well...
Degustator: it's a statement based on working with most currently available architectures (Cypress, Cayman, Tahiti, Fermi). That may seem a bit surrealistic to you, but these chips do serve purposes other than forum debates and giving reviewers means to generate hits through bar-charts. When Kepler is out, I'll talk about that - unlike you, I deeply resent talking about things I have no knowledge of (did my fair share of it, managed the difficult process of growing up I guess). This last point brings me to a small suggestion I'd like to make: you either shape up your discours or I'll send you away until a few months after Kepler launches, so that the noise level is reduced. It's only fair that I warn you before doing it. I'm convinced that you're capable of shaping up and having a non pure noise contribution, I hope you're willing to do it.
DegustatoR
Veteran
None of currently avialable architectures can be considered a direct comtetitor to GCN and no current chip is a direct competitor of Tahiti. If you with all your supposed knowledge of currently available architectures don't understand even that then I don't see any point in argueing with you any longer. And if my contribution is considered a noise by you then fine, I'll won't post anything here anymore. Keep thinking that Tahiti is the best that can be done on 28nms solely because it's better than a two years old architectures from competition.
This forum is turning into AMD's marketing tool fast.
Well, at least I tried. I don't think that "the only competitors are those that I like to put in a competition, too bad the one that I like isn't out" is a reasonable stance, and it's pretty clear you won't make an effort to be less of a noise generator. That's that I guess, so now let's get back to technical discussion.
http://www.hardocp.com/news/2012/01/25/nvidia_accused_using_marketing_shills/
nVidia accused of using marketing shills, $300 GK104 being part of this false marketing - ChipHell banned all of the spreaders
nVidia accused of using marketing shills, $300 GK104 being part of this false marketing - ChipHell banned all of the spreaders
I'd guess it's as simple as "new users, all spreading same info, IP from same company IP-space, possibly one known for doing similar tricks before
Why do you say that? The die size bogeyman is still very much alive in this thread at least.
Btw, with the exception of GK107 in the drivers have any other code names actually showed up in nVidia material? Seems like they're all made up based on Fermi nomenclature.
NVIDIA released and even updated more accurately a marketing slide mentioning sustainable DP FLOPs per Watt comparing Fermi, Kepler and Maxwell Teslas.
That for me is enough indication that they don't intend anytime soon to abandon the HPC market (rather the contrary) and the next best theory that they might that early develop a HPC only core sounds like nonsense to me too. Margins are high in that market but volumes are abysmally low, meaning it at least sounds irrational to me that they could get the R&D investment for a HPC chip just out of the HPC market. If the latter should grow enough in the less foreseeable future then yes it would be a possibility, but how much can it really grow after all especially since Intel, AMD and NVIDIA operate or intend to operate in it?
Anyway if anyone would had asked me for a guess not too long ago about different upcoming NV architectures I would had said that the two level/register cache ideas are for Maxwell and hotclocks gone for Echelon/Einstein. Since there's a very high possibility that hotclocks are gone in Kepler already (I'd be VERY surprised if they're still there with all the recent rumor mongering) I am asking myself again and again if the caching related changes are also already in Kepler.
The ever repeating rumors want GK104 to scratch Tahiti's back and the first question mark that pops in my mind how it's possible since the first most likely has less bandwidth. Above theory at least makes some sense (in the contrary case I'd love to stand corrected). Here's some food for thought for discussions on a highly speculative basis, since I'm also tired of the speculations about SKU names. They're going to call it Borgo Sphinxter 6969 (not really but who cares?) and no it doesn't have any XDR2
That for me is enough indication that they don't intend anytime soon to abandon the HPC market (rather the contrary) and the next best theory that they might that early develop a HPC only core sounds like nonsense to me too. Margins are high in that market but volumes are abysmally low, meaning it at least sounds irrational to me that they could get the R&D investment for a HPC chip just out of the HPC market. If the latter should grow enough in the less foreseeable future then yes it would be a possibility, but how much can it really grow after all especially since Intel, AMD and NVIDIA operate or intend to operate in it?
Anyway if anyone would had asked me for a guess not too long ago about different upcoming NV architectures I would had said that the two level/register cache ideas are for Maxwell and hotclocks gone for Echelon/Einstein. Since there's a very high possibility that hotclocks are gone in Kepler already (I'd be VERY surprised if they're still there with all the recent rumor mongering) I am asking myself again and again if the caching related changes are also already in Kepler.
The ever repeating rumors want GK104 to scratch Tahiti's back and the first question mark that pops in my mind how it's possible since the first most likely has less bandwidth. Above theory at least makes some sense (in the contrary case I'd love to stand corrected). Here's some food for thought for discussions on a highly speculative basis, since I'm also tired of the speculations about SKU names. They're going to call it Borgo Sphinxter 6969 (not really but who cares?) and no it doesn't have any XDR2
