NVIDIA GTX 680 2GB Review Comment Thread

[IRQ Conflict]

Originally Posted by SKYMTL

Why should we include that? Synthetic benches do nothing to indicate in game performance.

Just curious. Considering they reduced their "PolyMorph Engine count". But no need. I found what I was looking for. Seems they have really done some good things with this card. Stellar review SKY, thanks!

[esperasoppa]

SKY, any chance we see GK110 this year?

[Lsv]

I'm surprised at the 680 getting really damn close to 590 performance. I don't remember a card that come this close to 100% faster than it's predecessor.

Edit - also I noticed that HWC covered TXAA info while some other sites did not. I wish TXAA was driver enabled though :(

[SKYMTL]

Originally Posted by esperasoppa

SKY, any chance we see GK110 this year?

I'll plead the fifth on this one. Especially considering it all boils down to how successful TSMC's 28nm process is. If they continue having problems, AMD will continue having major supply issues with the HD 7970 and NVIDIA will try to avoid launching a larger chip.

[YoungMan]

HEHEH, I really like this move by Nvidia. I hope nobody buys any 7970s or 7950s until they drop the prices significantly.

[draemn]

Originally Posted by chrisk

Great review as always. Had to wait until just now to read more than the conclusion.

For those looking for folding numbers, this will be very interesting. Conpute functions are drastically changed, and there is a conpute bench out there (Luxmark) that is actually not very promising on that front. Actually, the 580 beat the 680. Hopefully its just a case of the drivers not being optimized for the benchmark used, and it is an OpenCL bench so there is hope. But when asked about it, nVidia seems to have said an update is coming, but they also said Kepler was not focusing on folding, but Physix instead.....

Will wait for folding numbers on this one big time.

Everyone has been saying that this is the one big weak area is compute. The 7970 has no problem claiming a win in one area...

All the people posting "hate nvidia" comments can have their little win.

[chrisk]

Originally Posted by draemn

Everyone has been saying that this is the one big weak area is compute. The 7970 has no problem claiming a win in one area...

All the people posting "hate nvidia" comments can have their little win.

Will have to wait and see if nVidia puts anything in their drivers and if they can help Stanford in optimizing a new core.

Without knowing much about Kepler, I took a quick look at the specs and saw 3 times the cores and nearly jumped out of my seat but that enthusiasm disappeared quickly when I actually paid closer attention.

Just have to wait and see, but for hardcore folders I would not head out right away and grab one of these for folding.

Who knows...maybe the 580 folds as well as the 680, making the 580s and 560s cheaper at least. No doubt that the 680s are gaming beasts, just as the 7970s are.

[SKYMTL]

LuxMark uses OpenCL and is buggy as hell. I wouldn't trust its numbers at all which is why you don't see it included in our tests.

If you want real compute versus compute performance from one generation to another, look to CUDA-based app benchmarks. In addition, please remember that this core was originally supposed to replace the GF114, which wasn't a compute-focused architecture to begin with....

[draemn]

The way the last few official drivers have broken GPU folding... I'm not holding my breath for drivers to make the 680 a good folder suddenly :/

[Loki76]

I have to say that this can't be the real flagship card for Nvidia.

If it is, it's extremely lackluster in performance, sure it has excellent performance/watt, but not overall performance.

Nvidia was saying how they are not impressed with AMD's new architecture and core, yet this supposed flagship can at best beat it with ~12%.

Also it has exactly the same bandwidth of previous generation, and ONLY has that because it overclocks the memory higher, if they didn't it would have a LOWER bandwidth.

Here comes the real meat of why this is a serious UNDER-performer compared to previous "new" generation architectures.

The 480 Fermi flagship card, beats the previous generation flagship card the 285 (an actual refresh card from the flagship 280) by a whopping 66% on average.

THIS card barely beats IT'S previous generation card (stated in this review) by 36%.

I am actually perplexed by why Hardware Canucks reviewer doesn't delve deeper into this, why doesn't he speculate why they have such a low bandwidth and a step-back in bus speed from 384 to 256-bit?

Nvidia has said all along that this Kepler is one of the best architectures they have EVER made in wins and performance etc.
How is this not screaming right in the face of everyone, especially PROFESSIONAL reviewers that something doesn't add up.

Seriously, 36% vs 66%, that is almost DOUBLE the performance difference.

What about the fact that this card was supposedly named GTX670ti, IE NOT a flagship naming for the card.
Also, Die-size, bandwidth, bus-speed all screams of mainstream performance card and NOT top-end card.

I would like to have seen the reviewer addressing these things and actually pose these questions in his review.

Also beating AMD's TOP card by 12% when AMD themselves can't even beat the previous generation GTX580 by more than 20-25%, that doesn't spell YAY what a success, it spells pretty weak performance.

Also several tests/reviews have shown that the 7970 sits on par (basically) with the GTX680 when they OC the cards (both of them).
Then you have the multiple monitor evidence, here you have a card that beats AMD by 12% yet loses out in multiple monitor tests.

Yet you still call this a "flagship" card.

Has everyone lost their understanding of what that means? It means the best they have to offer.

This is nothing but THIS generations "GTX 560ti" card, renamed to GTX680.
The price they ask for this card is horrendous it should be priced at $399, where is the GK110 REAL flagship card? That one should cost $600 and it's little brother (aka 570 version equivalent of this generation) should cost $500.

This is nothing but Nvidia passing off the mid/lower top end card as THE top-end card and charge accordingly.

Reviewers show some responsibility and acknowledge these things.
How can you call this card a "win" when it offers half the performance increase compared to almost ALL other architectures Nivida has done in previous generations between them.

Also, as a Disclaimer I'm an Nividia Fanboy, I have only ever bought Nvidia cards since over a decade and a half.

Starting with the very first Geforce card.
I'm getting another Nvidia card soon since I am upgrading, but this card rubs me the wrong way, and I feel annoyed by all reviewers that completely ignore this Big elephant in the room of all the things I said above.

You could have said that it is awesome performance/watt change in their architecture, but where is the real performance increase between generations? Where is the larger bandwidth.

No one in their right mind could ever think this is the best Nvidia has to offer, that they would put all their cards (pun) in one basket and hope to beat AMD with a 294mm die size chip, with a 256-bit bus with the same bandwidth as a GTX580.

Just like I felt super buyers remorse buying the GTX470 right away and then 5 months later the GTX580/570 coming out with same price but much more performance etc etc.

Can anyone here begin to imagine the increase of their REAL flagship card GK110 that might have a 512-bit bus and 50% MORE performance over the GTX680.
THEN people will realize that Nivdia pulled the wool over the face.

Long rant I know, but seriously where is the journalism here?