[Sxotty]

People might be afraid of it happening actually as opposed to hoping that one of them will disappear. I am much more worried about Nvidia disappearing since Intel must have AMD around.

[hatter]

From Charlie's twitter thread:

More on this soon, there is a BIG secret in the Kepler line. It will take a long article though, so not RSN.

And in response to:

You've said on the forums at SA that the 7970 will be better in performance than the GK104 but what about the 580 to the GK104?

He said:

It is really nuanced. I am working on this now, all I can say is stay tuned.

[trinibwoy]

Originally Posted by Rangers

Huh? You do realize 6970 is priced like $330 vs $550 for GTX 580, and is more like an equal to GTX 570 in performance?

The 580 is only 15% faster than a 6970, which has a $370 MSRP btw.

You seem to be vastly overrating 6970 performance for some reason.

Nope you're underrating it.

Nvidia is magically able to be teh more profitable with the big dice.

What's so magical about stating facts? Is that not the case?

[Alexko]

Originally Posted by trinibwoy

The 580 is only 15% faster than a 6970, which has a $370 MSRP btw.

Newegg currently has one for $304.99 with Rebate + $7.87 Shipping, though. Tahiti has more memory, but is a bit smaller, albeit on a more expensive process. All in all, and even though I'm sure AMD would love to keep it closer to $500, it should remain quite profitable around $300~350.

[Rangers]

Originally Posted by trinibwoy

The 580 is only 15% faster than a 6970, which has a $370 MSRP btw.



Nope you're underrating it.



What's so magical about stating facts? Is that not the case?

So $370 MSRP (if even true) that streets for $330 is supposed to be a huge difference from $300? Why not just admit you were wrong?

Edit: checking newegg it's true there's one model for 325/300, but most are 349 before rebate. I'm willing to be generous and even concede a 350 price, I dont think that changes the debate much.

I'm underrating 6970's performance by stating it's less than 580 and more of a 570 competitor? Nope, fraid I'm exactly right actually.

What's magical is basically stating that die size is meaningless to cost, which is basically what you have tried to do in the past in a roundabout, obfuscating way. As I said, in this thread it's funny how die size became important to cost again all that sudden, now that it's presumed GK104 may have a tiny die or some type of die size advantage. Just funny that's all.

[trinibwoy]

Originally Posted by Rangers

So $370 MSRP (if even true) that streets for $330 is supposed to be a huge difference from $300? Why not just admit you were wrong?

You must be arguing with yourself. I stated that it's been a long time since AMD has had to fight nVidia's second tier chip with their largest. How is the fact the 6970 matches the 570 and competes with the 580 proving me wrong exactly? I don't even know what point you're making....

I'm underrating 6970's performance by stating it's less than 580 and more of a 570 competitor? Nope, fraid I'm exactly right actually.

We're discussing a possible Tahiti vs GK104 scenario. How is the current situation of Cayman vs GF110 representative? Care to explain that?

What's magical is basically stating that die size is meaningless to cost, which is basically what you have tried to do

You really should spend more time reading and less time reacting. I said nVidia's costs have nothing to do with AMD's profit potential so a die size comparison is useless. nVidia is willing and able to sell its larger dies at similar prices to AMD's smaller ones, they've done so for years now. i.e. die size and performance parity is certainly not a sign of fair weather for AMD.

Not sure how else to say it to make it clearer.

[Sunrise]

<deleted>

[iMacmatician]

For GK104? GK110? Something else?

[psolord]

Hey guys.

I've been thinking....could GK104 actually come with a 384bit bus, despite the 2GB framebuffer?

I could be wrong, but hasn't this already been done, with the GTX 550 Ti having 192bit bus and 1GB framebuffer?

[AnarchX]

The chip is quite small, and has 8 GDDR5 chips meaning a 256-bit bus/2GB standard

http://semiaccurate.com/2012/01/19/n...-clear-winner/

384-bit = 12 chips

[DegustatoR]

Originally Posted by no-X

I belive the fastest GPU isn't going to be launched anytime soon.

Ah but it will. Although it depends on your definition of "soon" I guess.

[psolord]

Originally Posted by AnarchX

http://semiaccurate.com/2012/01/19/n...-clear-winner/

384-bit = 12 chips

I missed that info. Thanks.

[entity279]

Originally Posted by DegustatoR

Ah but it will. Although it depends on your definition of "soon" I guess.

Or depends on yours

[Man from Atlantis]

NVIDIA - "We expected more from Radeon HD 7900"

[Alexko]

Reminds me of this: http://forum.beyond3d.com/showpost.p...postcount=2339

[DarthShader]

Originally Posted by AnarchX

384-bit = 12 chips

What about the GTX550Ti with a 192 bit bus and 1GB Ram? Same trick could be used here.

[AnarchX]

Originally Posted by DarthShader

What about the GTX550Ti with a 192 bit bus and 1GB Ram? Same trick could be used here.

550Ti has 6 memory chips (4x128MiB + 2x256MiB). Minimum for 384-bit bus are 12 chips (12*32-bit ).

[Gipsel]

Originally Posted by Man from Atlantis

NVIDIA - "We expected more from Radeon HD 7900"

What did they expect? I would be genuinely suprised if they thought AMD's 365mm² chip were going to beat their own ~500mm² GPU on the same process. Of course Kepler "will have no problem taking down Radeon HD 7900". If nV would think differently, they would expect a massive failure on their side.

[psolord]

Originally Posted by DarthShader

What about the GTX550Ti with a 192 bit bus and 1GB Ram? Same trick could be used here.

I was thinking the same thing, without being aware that we knew the number of the chips as well.

If we only knew that it was 2GBs, we could have a 384bit bus with two of those.



edit: I should refresh more often. oh well.

[Bouncing Zabaglione Bros.]

Originally Posted by Man from Atlantis

NVIDIA - "We expected more from Radeon HD 7900"

What, a massive chip clocked and volted to the edge of it's life that draws 400 watts and sometimes explodes, all in order to win benchmarks for the small number you ship?

[trinibwoy]

Originally Posted by Gipsel

What did they expect? I would be genuinely suprised if they thought AMD's 365mm² chip were going to beat their own ~500mm² GPU on the same process. Of course Kepler "will have no problem taking down Radeon HD 7900". If nV would think differently, they would expect a massive failure on their side.

They are of course just bullshitting because they have nothing to show now but why do you assume they're comparing it to their ~500mm² GPU?

[AlexV]

Originally Posted by Gipsel

What did they expect? I would be genuinely suprised if they thought AMD's 365mm² chip were going to beat their own ~500mm² GPU on the same process. Of course Kepler "will have no problem taking down Radeon HD 7900". If nV would think differently, they would expect a massive failure on their side.

This is marketing gaga, why people take it at face value is beyond me (assuming rationality and no "Oh my ceilingcat, IHV X makes me cream my underwear just by existing" nonsense). Remember when Cypress launched? Some NV twitter feed was something along the lines of "is that all they've got? That's weak, we'll have hugely better stuff really soon now!!!!". I think we all know how that panned out (as a note, all those tempted to go on the "yes, it clearly resulted in the fastest chip with epic compute and great tessellation", I'm familiar with that rhethoric thank you). Remember the JF_AMD chap and BD, and the buildup? Yeah, we know how that turned out as well (I'm familiar with his weak excuse for that one too). I don't see any difference here - any public facing fellow worth his salt won't be allowed to say anything else.

[GZ007]

AMD lost a lot of efficieny with CGN architecture to catch up with Fermi. With Fermi tweaks and 28nm shrink kepler could be much better than the overpriced 6970 - 7970 performance jump.

[CarstenS]

Honestly, guys. What do you expect Nvidia to say other than that?

"Oh, godsdamnit, we need to cull GK104 and shell out GK110 as soon as it's ready for half the price we've intended to?"

[AlexV]

Originally Posted by GZ007

AMD lost a lot of efficieny with CGN architecture to catch up with Fermi. With Fermi tweaks and 28nm shrink kepler could be much better than the overpriced 6970 - 7970 performance jump.

Random statement is random, pulled out of thin air. Are we basing this on how many Crysis-fps it gets in site x's set of bar-charts? There's a bit more to Tahiti than has been ascertained into the first waves of coverage. Tahiti is a very nice chip to work with, and GCN looks quite solid architecturally.