GTX680 pics

[Mopetar]

I thought that someone in another thread posted information about that bar graph being fan made / fake. The forum posts for the linked article point out that the chart is baed off an older leak that was of questionable validity itself, so I'd take that chart with a grain of salt.

Also we already know that the cores are less powerful than their previous CUDA cores. You can check out the thread about the 640M which has some good information in it. However, it really doesn't matter so long as they've added enough cores to offset the relative difference between the two architectures. Napkin math in that thread suggests that Kepler will do pretty well.

 

[Dkcode]

Does this look like it will take current after market coolers for the 570/580 range?

 

[railven]

I thought that someone in another thread posted information about that bar graph being fan made / fake. The forum posts for the linked article point out that the chart is baed off an older leak that was of questionable validity itself, so I'd take that chart with a grain of salt.

Yep, in the know locked S|A Kepler thread, it was pointed out that chart was running around in January during the Chiphell fiasco.

 

[railven]

Does this look like it will take current after market coolers for the 570/580 range?

If that weird power connector makes it to the final version (which is visible in these pics with shroud) I'd have to guess your options would be limited.

It is more than twice the height of normal connectors if I had to guess. They are stacked but have a gap to accomodate the plugs.

 

[skipsneeky2]

Well, CUDA cores are only one part of the equation, there are many other factors that affect a GPU's overall performance.

That said, there is no way these are 1:1 as powerful as previous CUDA cores as the shader power would be insane and would probably hit 150+ GTexel/s..

Guess we will find out,but yup 3 times the cuda cores,it would have to be a monster and i have always been more of a nvidia guy but i ended up pulling the trigger on a 7970 but if this thing is boss,i would switch back just maybe.:biggrin:

 

[thilanliyan]

OMG MOAR PICS OF THE POWER CONNECTORS!!! Wutwut this is exactly the kind of info we need! And thankfully I don't see wood anywhere so we know it's real!

Seriously though, wake me up when we have some actual performance numbers. Hurry up and release already nV!! Your loyal nV fans are dying for something to argue about (although we here at VC&G can argue about pretty much anything lol), ATI fans are looking for pricedrops on 7 series cards, and those in the middle are looking for the best bang for buck which will hopefully come when there's competition.

 

[gorobei]

I assume people have seen this:



Edit: a few people below suggest this is a fake. But if this is indeed around $500-600, it better have performance like this.

generally less impressive looking if you use a proper bar graph rather than truncated.

 

[f1sherman]

generally less impressive looking if you use a proper bar graph rather than truncated.

particularly if you are not acquainted with the fine art of reading decimals in Indo-Arabic notation

 

[Mopetar]

Considering that it's probably not even real numbers, does it really matter what kind of presentation style is used?

Also, that's a fairly standard format that's used by both companies in their marketing, so I don't think it's a big deal even if the numbers weren't fabrications.

 

[chronochime]

I assume people have seen this:

I'm pretty confident that these numbers are correct. It wouldn't take much for Nvidia to counter the 7970, which was only 20% faster than the GTX580.

Edit: a few people below suggest this is a fake. But if this is indeed around $500-600, it better have performance like this.

People already mentioned in other forum that the graph is just based on that other (fake) chart, so this is nothing new. And that pic of the retail card might be fake as well, as the letters in Geforce don't have different font sizes like how Nvidia usually shows that word...

 

[DeeJayeS]

wtb real info from NV, pst

 

[Imouto]

generally less impressive looking if you use a proper bar graph rather than truncated.

Dunno what's doing the 7950 there anyway.

 

[taltamir]

Because NVIDIA. GTX580 was 250W, GTX 480 was 250W

which they earned a lot of flak for, so maybe they improved now.

Anyways, considering that laptops with kepler are on sale now in asia since two days ago, I suspect the rumors of a desktop part imminent makes sense too.

 

[VulgarDisplay]

You can tell that chart is fake just by looking at the civ 5 numbers. Nvidia cards run Civ 5 far better than AMD, and somehow the 7970 beats the 680 in that game. Don't think so.

 

[taltamir]

You can tell that chart is fake just by looking at the civ 5 numbers. Nvidia cards run Civ 5 far better than AMD, and somehow the 7970 beats the 680 in that game. Don't think so.

Different architecture. The 7970 is extremely similar to fermi in architecture. While keplar, well we don't know what they changed but it is possible that the changes are such that it does worse in civ 5.

Although, I don't trust these type of charts... I want FPS measurements from reliable trustworthy reviewers like anandtech not supposedly leaked marketing nonsense charts (1X comparisons)

 

[BallaTheFeared]

Different architecture. The 7970 is extremely similar to fermi in architecture. While keplar, well we don't know what they changed but it is possible that the changes are such that it does worse in civ 5.

Although, I don't trust these type of charts... I want FPS measurements from reliable trustworthy reviewers like anandtech not supposedly leaked marketing nonsense charts (1X comparisons)

uArch wouldn't matter, Nvidia's drivers are what makes the difference. Even the 6970 would've beat the 580 in Civ V without Nvidia's A game on drivers. In fact it did, until Nvidia released drivers with support for a key DX11 feature that not at all surprisingly in our console port world is only supported by one game.

Humm, not to say uArch wouldn't matter but CivV is texture heavy, that's why AMD cards had an advantage in it prior to the Nvidia driver release. If anything this new design from Nvidia should improve the texture fillrate of their cards which was the bottleneck most likely for the 580.

 

[Rvenger]

Dunno what's doing the 7950 there anyway.

Fake... But I hope it somewhat holds true so we can see some nice price drops on AMD.

 

[tviceman]

I assume people have seen this:



Source: http://www.techpowerup.com/162305/GeForce-GTX-680-Up-To-40-Faster-Than-Radeon-HD-7970-NVIDIA.html

I'm pretty confident that these numbers are correct. It wouldn't take much for Nvidia to counter the 7970, which was only 20% faster than the GTX580.

Edit: a few people below suggest this is a fake. But if this is indeed around $500-600, it better have performance like this.

It is fake. It was lifted from this chart that was floating around MONTHS ago:

Battlefield, Call of Duty, and Civ V do not score anywhere close to that high on either the gtx580 or hd7970. Also, Dirt 3 runs way better on both gtx580 and hd7970.

 

[tviceman]

Because NVIDIA. GTX580 was 250W, GTX 480 was 250W

So how much power do you think a 295mm^2 chip with 2 6-pin power connectors on 28nm will consume?

 

[Barfo]

Where are the special, heat dissipating wood screws?

 

[BallaTheFeared]

/slapknee

 

[ArchAngel777]

I assume people have seen this:



Source: http://www.techpowerup.com/162305/GeForce-GTX-680-Up-To-40-Faster-Than-Radeon-HD-7970-NVIDIA.html

I'm pretty confident that these numbers are correct. It wouldn't take much for Nvidia to counter the 7970, which was only 20% faster than the GTX580.

Edit: a few people below suggest this is a fake. But if this is indeed around $500-600, it better have performance like this.

I abhor graphs that don't start at zero. What a way to mislead the majority of people who failed math and statistics. Then again, maybe they deserve it.

 

[Soccerman06]

So how much power do you think a 295mm^2 chip with 2 6-pin power connectors on 28nm will consume?

Max power supposedly with pci-e limitation (which has never been exceeded :hmm is 75w so thats 225w total.

 

[Keysplayr]

/slapknee

Could you imagine if this chart was accurate? I'd head for a bunker.

 

[MrTeal]

/slapknee

What's the source on that?
Really, die size should be pretty locked down from the pictures released at this point. If Tahiti really does have 21% more transistors and 14% more power dissipation for the same load, even a performance tie would be a big win for nVidia.

If the GTX680 performs better than the 7970, that would be such a huge disparity that AMD would be screwed this generation. Even at price parity the GTX680 would be a better choice as it performs better, and costs for nVidia would be lower so margins would be higher.

That being said, for the 680 to have 10% more performance than a 7970, it would need to have 36.6% better perf/mm^2 than Tahiti. That's a bit hard to believe. I'd say it's much more likely that GK104 performs similar to or a little below the 7970, on average.