Nvidia Kepler vs AMD GCN has a clear winner

http://forums.anandtech.com/showthread.php?t=2220743

Poll: How did NVidia get Kepler to beat Southern Islands?

Has Nvidia got a winner around the corner? I can't believe the Semi Accurate article linked first is wholly accurate!

Can 256 memory bus beat a 384 one?

Read that linked on another forum a couple of days ago and i'll believe it when i see some proper benchmarks. That being said, they probably will win the top end as they usually do, not that i can ever afford those cards. I'm more interested on what both companies will do with their mid range ones ie 660 and 7950.

Stopped believing these kind of "news" years ago.

Until cards hit shelves and tech sites then all of those info may be simply not true.

Remember how Bulldozer supposed to be 'oh so great' CPU?

That's Charlie. He's always for AMD and more recently the 7970 being awesome. I think he either got hacked or is looking for websites that steal his information.

Charlie has some real sources, and has broken some major stories in the past, so this isn't equivalent to a story from BSN or Fudzilla, let alone some random person with Internet access.

The allegations of Charlie being biased against Nvidia were always overblown.  Two years ago, he was jumping up and down screaming to anyone who would listen that Fermi was hot, late, broken, and various other bad things.  People accused him of hating Nvidia over it.  But he was right.

That said, the story isn't very specific, and Charlie doesn't do mild-mannered.  If Kepler beats GCN by 5% in performance per watt and performance per mm^2, he'd probably claim that the story was right.

As for the technical side, Charlie is also claiming that the top Kepler chip was cancelled.  So GK104 versus Tahiti is the successor to GF104 versus Cypress, or GF114 versus Cayman.

It's hard to imagine a 256-bit GDDR5 GK104 beating a high clocked 384-bit Tahiti (which is already meaningfully limited by memory bandwidth) in raw performance unless Nvidia has found some miraculous way to reduce memory bandwidth requirements for a given level of performance.  But if GK104 can get 90% of the performance of Tahiti for 70% of the cost of building it, is that a win for Kepler?  I'd say so, for the same reasons that I'd regard the Radeon HD 4000 series as having beaten the GeForce 200 series.

Let's also note that final performance numbers aren't yet knowable.  GCN probably has a fair bit of improvements from drivers coming over the coming months.  Kepler probably doesn't even have stable drivers for many games yet, so that should have huge improvements coming, too.

It's likely that Kepler would be a bigger improvement over Fermi than GCN is over Northern Islands, for the simple reason that it's easier to improve on your last effort when your last one was pretty bad.

Finally, let's not forget that this is Nvidia's first chance at coming up with a new architecture after seeing how badly AMD beat Nvidia with RV770 (Radeon HD 4800 series).  When Nvidia was designing Fermi, they saw how badly G80 (GeForce 8800) had beaten R600 (Radeon HD 2900), and had reason to believe that continuing along the lines of what they had was the way to go.  Presumably Nvidia was able to learn from RV770's approach, just like AMD was presumably able to learn something from G80 years earlier.

ATI and Nvidia have been fighting hard and one or the other company have had the lead.

ATI have recently had slightly faster cards and that it can turn again is not unlikely but I think we shouldn't be sure one way or another yet.

I will not really say that one or the other company is better. I can admit that most of the computers I build for buddies have Nvidia, but that is frankly because their drivers are easier to download and my buddies are rather incompetent with stuff like that (which is why I get a case of beer to build their computers instead of them doing it themselves). Most computers I build with ATI cards are the budget rigs, Nvidianever been good on low budget cards.

Certain cards are however better than others.

I am looking forward to see how next gen cards will really measure against eachother in reality. :)

Doesn't matter to me who wins!  I've always loved and will purchase an nVidia GPU again.  The great thing this round however, is that I've got plenty of cash saved up to get one of the new cards as soon as they're available.  Course I'll probably wait until proper benchmarks prove that they're substantially better than the EVGA GTX 580 I currently own. 

I'd also like to upgrade my CPU to one of the new Ivy Bridge models.  I'm running an older i7 K875 Lynnfield, that I originally bought to overclock, but paired with the 580 at stock speeds, there isn't a single game currently available that gives me any trouble.  I easily get 45-80+ FPS in everything...{makes me wonder why I should even consider upgrading then, but I do like having an overpowered system...heh   And it does help a great deal when running MMOs with tons of players/enemies on-screen.  GW2 WvWvW and those epicly~massive boss fights!}

---

Originally posted by Khrymson
Doesn't matter to me who wins!  I've always loved and will purchase an nVidia GPU again.  The great thing this round however, is that I've got plenty of cash saved up to get one of the new cards as soon as they're available.  Course I'll probably wait until proper benchmarks prove that they're substantially better than the EVGA GTX 580 I currently own. 

---

That's exactly the type of owner they love to have. /salute

Another interesting twist.

http://hardforum.com/showthread.php?t=1668756

Apparently a major Chinese-language tech site is claiming that the rumors of Kepler being awesome are coming exclusively from paid Nvidia shills, that they can prove this, and that they have banned all such accounts from their site.  It's also claimed that these were the sources behind Charlie's post on SemiAccurate, though he'd probably deny that (edit: and now has).

My take on it is this:  if GK104 is as good as claimed, then the day it launches, GF110 is dead for consumer use and will have to be priced substantially cheaper than GK104.  A GeForce GTX 580 selling for $250 or a GTX 570 for $200 (as they'd have to if GK104 is $300 and as good as claimed) would probably mean Nvidia losing money on every card sold.  Remember that GF110 is a huge, expensive chip, and there are a lot of other expensive components that go into making a high-powered card work.

Thus, Nvidia wouldn't want to still have a bunch of GF110 chips laying around once GK104 arrives, and would discontinue GF110 well in advance.  Don't believe that a GK104 that beats GF110 selling for $300 for the top bin at retail is just around the corner until the GTX 570 and GTX 580 are either mostly disappearing or available at deeply discounted prices.  Currently New Egg has tons of both of them in stock, and the GTX 570s are actually more expensive than they were a few months ago.

Indeed, even if GK104 is as good as claimed, Nvidia wouldn't necessarily want that leaking out so early.  That's the old Adam Osbourne effect:  hype your next generation too much and sales of your present generation plummet.  Note that Nvidia fanboys are the ones most likely to wait, as AMD fanboys would just take a Radeon card regardless.

There are a lot of different scenarios that would be consistent with what we see so far.  One is that GK104 really is as good as claimed, the GTX 570 and GTX 580 have been discontinued for a while, and we're going to start seeing SKUs disappearing soon.  Another is that GK104 is still a long way off, coming perhaps in the latter half of 2012 rather than March or April.  Yet another is that it's a great chip, but the top bin will sell at $500 or so, not $300.  Or maybe it's just not that great of a chip and as broken as Fermi was.

I will say that GK104 beating GF110 is highly plausible.  As bad as Fermi was, GF104 managed to roughly keep pace with GT200b, the top chip of the previous generation.  With Fermi, Nvidia seems to have moved to a model of an upper-midrange chip that cuts out all of the GPGPU stuff (so as to be cheaper to produce) being not that far behind the top chip in gaming performance, and it's highly plausible that they'd continue that with Kepler.

I read the shill article and thought "Hmm, they got caught."

It shouldn't surprise anyone that there are shills out there. I have my suspicions about more than a couple on this site even.

I'm sure AMD and nVidia and even all the 3rd party manufacturers (EVGA, Sapphire, etc) all have their own shills out there working - I have no doubt, they just haven't been caught yet. When they actually come out and say they are Company Representatives, I have a lot of respect for them. A few have even come on this forum.

When they try to sneak around and just bash one side, then it's underhanded, and unfortunately it's common practice. Belkin was caught red-handed paying people to post positive reviews for their products on Amazon. Several less scrupulous review sites have been caught accepting payment in return for positive reviews - sometimes in cash, most often times in "Free Review Samples".

This is why I encourage people to read, but not rely upon, reviews posted on sales sites. Anyone someone links some piece of hardware stating that it has 5 eggs, or all positive reviews, I keep thinking of Belkin, and now nVidia...

Might be my first nVidia in a while if its true.  An effecient card with some smart cores is just what my system needs.  The 7970 seems a little bloated to me.

BTW most Belkin is shit grade stuff.  Only good Belkin product I have ever seen is the N52. 

http://semiaccurate.com/2012/02/01/physics-hardware-makes-keplergk104-fast/

Now that sounds more like a Charlie article.  He's claiming that GK104 will beat Tahiti in some games, and lose to Pitcairn (the next chip down) in others, with performance varying wildly depending on whether the game uses some PhysX optimizations that Nvidia is working on.  And that includes running PhysX on the CPU, not just the GPU.

Presumably we'll just have to find out about performance when the cards launch.