Kepler Nvidia GeForce GTX 780

[SKYMTL]

You can see the file name by hovering your mouse over the pic, it has the game name like I stated. Whether it's just "allocation" or use is left for you or some other reviewer to test. I can only point to my own experiences where 6950CF simply didn't provide as good performance even though it's technically faster at lower resolutions.

Here is an easy (but still not 100% accurate) way of testing it:

When the in game memory amount is ABOVE the card's actual installed capacity, check the GPU usage. If it is still at 100% then it is likely the GPU core itself chugging along. If usage drops drastically (say to 75% and lower), then memory bottlenecking is a distinct possibility. This is because the GPU will throttle itself if the memory can't keep up with its dispatch requests.

[Tim]

Nice vid, really looks promising. Interesting to see the rumour mill now and see what was correct and what not.

bad "pills" for GTX680 mega machine overhyped:

- only 2GB VRAM
- DX11.0???
- Very Expensive
- Turbo boost for what?
- low rate in floating point compute....

what happened to NVIDIA?

Troll fail ha ha.

[SKYMTL]

what happened to NVIDIA?

Seriously! If all the rumors are to be believed, they've suddenly decided to beat AMD at efficiency AND performance. Geeze, what's this world coming to? /end sarcasm.

[Manicdan]

Seriously! If all the rumors are to be believed, they've suddenly decided to beat AMD at efficiency AND performance. Geeze, what's this world coming to? /end sarcasm.

sadly high prices, unless someone changes this.
im still spoiled from the 4xx0/2x0 series and cant see anything worth upgrading too for the price.

[Vicey]

I guess NVIDIA really has focused on power consumption and heat if it has just two 6 pin power connectors. When was the last time they launched their top end card with just that? 8800 GTX and GTX 285?

I'm really quite surprised to be honest. I think the memory is a little low for Surround Gaming (and he says in the video a single card supports that now) but I'm sure a 4GB model will show up eventually. Perhaps a GTX 685?

[kaktus1907]

it's been boring when you had no countdown


date: "march 22, 2012 14:00:00"
http://www.webhallen.com/hardvara/ka..._launch_event/

[dartaz]

GTX 680 2GB is better overall @1600p according to tom's hardware review. Who cares how much memory it has ?! What really matters is real word results.

As for compute performance, how many people seriously care about that ?! HD7970 might have advantage in this, but GTX 680 has advantage it other areas such lower power consumption, 4 display support and PhysX support. Not to mention that nvidia usually have better driver support.

Kepler is a revolutionary architecture. It performed better than what was expected. It beats AMD with smaller die size and lower power consumption. However, AMD fanboys will never admit this.

[paulbagz]

Kepler is a revolutionary architecture. It performed better than what was expected. It beats AMD with smaller die size and lower power consumption. However, AMD fanboys will never admit this.

No offense, but what's revolutionary about Kepler? =/

-PB

[GoldenTiger]

No offense, but what's revolutionary about Kepler? =/

-PB

Just about everything from what it looks like.

It appears to win on...

Price
Performance
Heat
Power consumption
Perf/watt
Perf/mm
Perf/price
Drivers
PhysX support
CUDA support for developers
Die size
Noise levels
AKA every conceivable measurable factor we know of.


We'll know for sure soon, but it's looking REALLY impressive as things stand now.

[flopper]

Just about everything from what it looks like.

It appears to win on...

Price
Performance
Heat
Power consumption
Perf/watt
Perf/mm
Perf/price
Drivers
PhysX support
CUDA support for developers
Die size
Noise levels
AKA every conceivable measurable factor we know of.


We'll know for sure soon, but it's looking REALLY impressive as things stand now.

forgot, Green, sexy, made by Nvidia, etc..

[[XC] Oj101]

Not much of that is exclusive to Kepler though. I think the main thing is the new TXAA.

[El Mano]

No offense, but what's revolutionary about Kepler? =/

-PB

It's the first time they release a new high end card with a smaller bus than actual high end cards.

[Kristoferr]

Will it outperform HD 7870 for perf/watt? I think it will be close one. I dont think that Kepler archidecture is far better. GTX 680 is just way better optimized for games than Tahiti XT is. Job well done for nvidia. Took some time, but they did it.

[Dami3n]

It's the first time they release a new high end card with a smaller bus than actual high end cards.

Nop. GTX 480 - 384 bits < GTX 285 - 512 bits

[GoldenTiger]

forgot, Green, sexy, made by Nvidia, etc..

I meant in comparison to the 7970.... do you have data showing my conclusions incorrect, because all known info shows it will almost definitely be right?

As far as "revolutionary", I'm hopeful adaptive vsync will be good, and TXAA... but we'll see what those bring to the table. The self-turbo thing looks intriguing as well.

[Manicdan]

Will it outperform HD 7870 for perf/watt? I think it will be close one. I dont think that Kepler archidecture is far better. GTX 680 is just way better optimized for games than Tahiti XT is. Job well done for nvidia. Took some time, but they did it.

this i agree with
i think the 256bit bus is just so much more effecient at typical review resolutions that it makes things look much better
lets not forget that the 7870 chip size is just over 200mm2,
the 7970 is only 24% faster (from TPU 1080p summary)
IF the 680 is 20% on top of that (1.24*1.2 = 1.48), then the 7870 and 680 are neck in neck at perf/mm2 because the kepler chip is about 1.5x larger

for perf/watt the 7800s are 10-20% higher than the 7900s, but if the 680 is pushing a wattage lower than a 7870 then its safe to say that its efficiency could be a solid 20% better than even the 7800s. the only problem is that on TPU the power draw for a 7900 is under well under 150w, and only 101w for a 7950. so maybe depending on how its tested this might look way different. for instance if you look at total system draw, the cpu might be pushed more as the gpu is able to get better frames, while other sites just read the power going through the 12v lines/pcie port.

[Andrew LB]

Not much of that is exclusive to Kepler though. I think the main thing is the new TXAA.

I dunno. Some might agree that for nVidia to not only take back the performance crown, but also to pivot and take direct aim at the one advantage AMD has relied upon for so long.... performance/watt/efficiency. I personally never thought nVidia would go that route because they've proven during this past decade that having the fastest chip on the market corresponds to big sales numbers.

I'll bet nVidia could have released a far faster chip but decided to hold back to see what AMD will respond with, and due to keeping clock speeds well within the silicon's capabilities, was able to make the necessary refinements to take on AMD not only in raw performance but while using the least amount of power as well.

[AliG]

this i agree with
i think the 256bit bus is just so much more effecient at typical review resolutions that it makes things look much better
lets not forget that the 7870 chip size is just over 200mm2,
the 7970 is only 24% faster (from TPU 1080p summary)
IF the 680 is 20% on top of that (1.24*1.2 = 1.48), then the 7870 and 680 are neck in neck at perf/mm2 because the kepler chip is about 1.5x larger

for perf/watt the 7800s are 10-20% higher than the 7900s, but if the 680 is pushing a wattage lower than a 7870 then its safe to say that its efficiency could be a solid 20% better than even the 7800s. the only problem is that on TPU the power draw for a 7900 is under well under 150w, and only 101w for a 7950. so maybe depending on how its tested this might look way different. for instance if you look at total system draw, the cpu might be pushed more as the gpu is able to get better frames, while other sites just read the power going through the 12v lines/pcie port.

One thing I do find very strange is that 7870 is a far more efficient chip than the 7970 in many respects. The performance scaling compared to die size increase and shader increase is rather poor imo. Makes me wonder if the 7970 is seriously handicapped by the 32 ROPs or perhaps even memory bus (though bandwidth wise it still is a big increase over the 6970)

[paulbagz]

Not much of that is exclusive to Kepler though. I think the main thing is the new TXAA.

That's what I was eluding too.

GPU Boost is new as well yes, but hardly revolutionary.

-PB

[clayton]

I'm more interested in mobile Kepler. GT 650m in Clevo 11" with rumored GTX 560m performance. AMD APU? Hah!

[prava]

Its funny it seems some websites won't be allowed to publish Kepler results even after NDA lift-off due to "non disclosed reasons". Those sites will only be allowed to publish tech-specs of Kepler but not a single benchmark, at least tomorrow.

Makes me wonder the reasons behind this...

[GoldenTiger]

Its funny it seems some websites won't be allowed to publish Kepler results even after NDA lift-off due to "non disclosed reasons". Those sites will only be allowed to publish tech-specs of Kepler but not a single benchmark, at least tomorrow.

Makes me wonder the reasons behind this...

Where are you hearing this?

[[XC] Oj101]

I dunno. Some might agree that for nVidia to not only take back the performance crown, but also to pivot and take direct aim at the one advantage AMD has relied upon for so long.... performance/watt/efficiency. I personally never thought nVidia would go that route because they've proven during this past decade that having the fastest chip on the market corresponds to big sales numbers.

I'll bet nVidia could have released a far faster chip but decided to hold back to see what AMD will respond with, and due to keeping clock speeds well within the silicon's capabilities, was able to make the necessary refinements to take on AMD not only in raw performance but while using the least amount of power as well.

I wouldn't call an increase in performance per watt revolutionary, it's natural progression.

[Russian]

Its funny it seems some websites won't be allowed to publish Kepler results even after NDA lift-off due to "non disclosed reasons". Those sites will only be allowed to publish tech-specs of Kepler but not a single benchmark, at least tomorrow.

Makes me wonder the reasons behind this...

What the hell are you talking about?

[Mad1723]

Well, one thing for sure. If Kepler is as good as it seems, I might have a worthy successor to my 5870. It served me well, but more memory is needed and BF3 kills it. Funny how I upgraded my video card for both Battlefield games in recent history (Bad Company 2, from a 4870 to a 5870 and now BF3, sounds like 5870 to GTX680!)