OFFICIAL KEPLER "GTX680" Reviews

[RussianSensation]

You mean 4-5 of the best current games from different genres.. if you aren't implying anything why do you feel the need to add your second line denigrating [H]? Is that really necessary?

Like 2 console ports with Mass Effect 3 and Deus Ex that can be maxed out on any GTX580 / HD6970?

Basically there are only 3 useful games in that entire review of the $600 Lightning card.

- No strategy games
- No MMOs
- No racing games
- No game that's not a Triple-A title to see how the cards perform in less popular games

I am pretty sure people spending $500+ on a card want to see benchmarks in 10-15 games at least.

If you are going to argue testing the best games of 2012, then we shouldn't even need to upgrade since Starcraft 2 expansion , Mass Effect 3 and Diablo 3 will easily be among the top 5 best PC games of 2012 - none of which need a new GPU upgrade.

Really the most demanding games were released 1-5 years ago: Crysis 1/2, Metro, Witcher 2, Dragon Age 2, Serious Sam 3, Anno 2070.

 

[f1sherman]

Going to the semiaccurate forums on a day like this is like going to redstate.com after the election of Obama: you feel very dirty, but it's totally worth it.

 

[blastingcap]

I want a card that also makes a decent tuna sandwich.

For that you will have to get a Matrox.

 

[Grooveriding]

Size comparison to HD6990:

Looks tiny, but I know the 6990 is huge. Is it smaller than a 480 you think or the same ? I hate the power connectors on this thing, why could they not of just put them side by side. Would like to know why the decision was made to do that ? Just to be different or is there some sort of logic to it. I can't see any good reason for it and can see a few that make it not a good thing. (will not fit a single-card form factor with a water block etc.)

 

[blastingcap]

Like 2 console ports with Mass Effect 3 and Deus Ex that can be maxed out on any GTX580 / HD6970?

Basically only 3 useful games in that entire review of the $600 Lightning card.

- No strategy games
- No MMOs
- No racing games

I am pretty sure people spending $500+ on a card want to see benchmarks in 10-15 games at least.

Seriously no one is going to buy a $500 card to max out Mass Effect 3 and Deus Ex.

To be fair, they opened up their selection to a vote a while back, so those game selections are at least partly the result of popular vote. And ME3 at Eyefinity resolutions might tell you something even if the 1080p stress case doesn't. As long as they include at least 2 heavy hitters, it shouldn't be a big deal.. it'd be like reviewing the GTX 280 using Crysis and whatever was no. 2 back then in terms of stress testing--every other game will be as good or better.

 

[blastingcap]

Looks tiny, but I know the 6990 is huge. Is it smaller than a 480 you think or the same ?

I heard that the GTX 680 is 10 inches long. I think most people here can handle that.

 

[Silverforce11]

Like 2 console ports with Mass Effect 3 and Deus Ex that can be maxed out on any GTX580 / HD6970?

Basically there are only 3 useful games in that entire review of the $600 Lightning card.

- No strategy games
- No MMOs
- No racing games
- No game that's not a Triple-A title to see how the cards perform in less popular games

I am pretty sure people spending $500+ on a card want to see benchmarks in 10-15 games at least.

If you are going to argue testing the best games of 2012, then we shouldn't even need to upgrade since Starcraft 2 expansion , Mass Effect 3 and Diablo 3 will easily be among the top 5 best PC games of 2012 - none of which need a new GPU ugprade.

Really the most demanding games were released 1-5 years ago: Crysis 1/2, Metro, Witcher 2, Dragon Age 2, Serious Sam 3, Anno 2070.

Your "demanding" games like Crysis 2, Metro, SS3 and DA2 are hardly being played so what does it matter? Are the majority of users going to upgrade to play those games? No.

[H] Bench games that are new and popular, its what the hardware will be used for, so i refute the statement by that obvious troller to claim [H] reviews shit games. If you want a bigger list of games, there are heaps of other review sites. Like TPU, Guru, and ABT, heaps of games.

 

[Silverforce11]

Looks tiny, but I know the 6990 is huge. Is it smaller than a 480 you think or the same ? I hate the power connectors on this thing, why could they not of just put them side by side. Would like to know why the decision was made to do that ? Just to be different or is there some sort of logic to it. I can't see any good reason for it and can see a few that make it not a good thing. (will not fit a single-card form factor with a water block etc.)

Looks around 9.5 inch, perfect fit for my rig. If performance holds up and it OCs well I'll grab one. If it OCs like crap, i'll save some $$ and get a 7870.

 

[RussianSensation]

To be fair, they opened up their selection to a vote a while back, so those game selections are at least partly the result of popular vote. And ME3 at Eyefinity resolutions might tell you something even if the 1080p stress case doesn't. As long as they include at least 2 heavy hitters, it shouldn't be a big deal.. it'd be like reviewing the GTX 280 using Crysis and whatever was no. 2 back then in terms of stress testing--every other game will be as good or better.

Ok but let's not use HardOCP as representative of overall performance then.

If someone doesn't play BF3, that Lightning review ends up testing just 2 useful games - Batman AC and SKYRIM.

I have no problem using HardOCP as 1 of 20 points of view but some posters here push it as having some ridiculous weight of 50%+ among all reviews. That's totally unreasonable since their game selection is extremely weak and a ton of gamers could care less about Eyefinity because of the bezels.

Also, if you care about Eyefinity, just check out Bit-Tech's review of HD7950 where HD7970 flopped hard on its face running 3x 1080P monitors. So really most people running modern games in Eyefinity already know they need 2 HD7950/7970s. They don't need reviews to tell them that.

 

[MrK6]

Like 2 console ports with Mass Effect 3 and Deus Ex that can be maxed out on any GTX580 / HD6970?

Basically there are only 3 useful games in that entire review of the $600 Lightning card.

- No strategy games
- No MMOs
- No racing games
- No game that's not a Triple-A title to see how the cards perform in less popular games

I am pretty sure people spending $500+ on a card want to see benchmarks in 10-15 games at least.

If you are going to argue testing the best games of 2012, then we shouldn't even need to upgrade since Starcraft 2 expansion , Mass Effect 3 and Diablo 3 will easily be among the top 5 best PC games of 2012 - none of which need a new GPU upgrade.

Really the most demanding games were released 1-5 years ago: Crysis 1/2, Metro, Witcher 2, Dragon Age 2, Serious Sam 3, Anno 2070.

Yet you just declared the GTX 680 clearly superior by referencing 3DMark11, and two console ports, Dirt 3 and Skyrim... interesting.

Looks around 9.5 inch, perfect fit for my rig. If performance holds up and it OCs well I'll grab one. If it OCs like crap, i'll save some $$ and get a 7870.

Only overclocks I've seen were from those supposed SLI tests at 1150MHz. I'm assuming that's on stock volts, which looks good so far.

 

[blastingcap]

Ok but let's not use HardOCP as representative of overall performance then.

If someone doesn't play BF3, that Lightning review ends up testing just 2 useful games - Batman AC and SKYRIM.

I have no problem using HardOCP as 1 of 20 points of view but some posters here push it as having some ridiculous weight of 50%+ among all reviews. That's totally unreasonable since their game selection is extremely weak and a ton of gamers could care less about Eyefinity because of the bezels.

Also, if you care about Eyefinity, just check out Bit-Tech's review of HD7950 where HD7970 flopped hard on its face running 3x 1080P monitors. So really most people running modern games in Eyefinity already know they need 2 HD7950/7970s. They don't need reviews to tell them that.

Dude you were the one who said a long time ago that nobody cares if you get 120 vs 180 fps in, say, Half Life 2. The only times GPU differences matter is when they affect framerates below the refresh rate of your monitor, or below some arbitrary "playable" limit like 30 or 24 fps (varies by person). So heavy games like BF3 ought to get more consideration. So you were saying that maybe [H] should just test the top few heaviest games and maybe a popular one or two, or something like that, IIRC. Have you since changed your mind?

Anyway I doubt we'll see huge differences among games nowadays with the new architectures; NV and AMD architectures have converged somewhat. There are still differences, but not like in the old days.

 

[RussianSensation]

Your "demanding" games like Crysis 2, Metro, SS3 and DA2 are hardly being played so what does it matter? Are the majority of users going to upgrade to play those games? No.

That's not why those games are tested. They are tested because they are the most demanding games. It's irrelevant if we have beaten them or not. Some people still have a backlog list of games too.

On the contrary, no one is buying $500 GPU to play Mass Effect 3. You can literally max out that game on a stock HD6950 with no overclocking.

[H] Bench games that are new and popular, its what the hardware will be used for, so i refute the statement by that obvious troller to claim [H] reviews shit games. If you want a bigger list of games, there are heaps of other review sites. Like TPU, Guru, and ABT, heaps of games.

Ya, so let's not use [H] to project overall performance of a card then. If people use [H] as just 1 data point, that's cool. But [H] should never be used to provide an overall recommendation on a card since it tests too few games to make such an absolute statement.

Websites such as Computerbase, GameGPU, Xbitlabs, Anandtech, Hardware Canucks, TPU do a far better job. Also same websites often come up with contradictory conclusions to HardOCP, exactly because they test so many games, which helps to remove outliers. HardOCP's review of HD7770 was the worst GPU review I've read in 10 years, easily.

Also like I said the most popular games on the PC space are MMOs, role-playing and Strategy games. So if you are going to argue based on popularity alone, then HardOCP reviews miss that mark by miles.

Dude you were the one who said a long time ago that nobody cares if you get 120 vs 180 fps in, say, Half Life 2. The only times GPU differences matter is when they affect framerates below the refresh rate of your monitor, or below some arbitrary "playable" limit like 30 or 24 fps. So heavy games like BF3 ought to get more consideration.

I haven't changed my mind at all 180 vs. 120 fps doesn't matter. But 60-80 fps avg matters because if you are getting 60 fps average, you are dipping well below 60. My point is they should include 10 games in their reviews if we are to take their "overall GPU recommendation" seriously. That's why their HD7770 review was a total joke. Across a wide variety of games, HD7770 is slow against an HD6870/GTX560 but at HardOCP the difference in playability was not any worse. If they include 10+ games, add a larger variety such as Anno 2070, Witcher 2 Uber sampling, Shogun 2, F1 2011 or Dirt 3 throw the most demanding PC games ever made - Crysis and Metro 2033, then we get a far better idea of the overall performance of 1 card vs. another. Then their overall GPU recommendation can be taken more seriously.

Otherwise, their reviews are only representative of the 5 games they test, not enough to make a general recommendation unless you spend 95% of your time playing BF3.

 

[Silverforce11]

That's not why those games are tested. They are tested because they are the most demanding games. It's irrelevant if we have beaten them or not. Some people still have a backlog list of games too.

On the contrary, no one is buying $500 GPU to play Mass Effect 3. You can literally max out that game on a stock HD6950 with no overclocking.



Ya, so let's not use [H] to project overall performance of a card then. If people use [H] as just 1 data point, that's cool. But [H] should never be used to provide an overall recommendation on a card since it tests too few games to make such an absolute statement.

Websites such as Computerbase, GameGPU, Xbitlabs, Anandtech, Hardware Canucks, TPU do a far better job. Also same websites often come up with contradictory conclusions to HardOCP, exactly because they test so many games, which helps to remove outliers. HardOCP's review of HD7770 was the worst GPU review I've read in 10 years, easily.

Also like I said the most popular games on the PC space are MMOs, role-playing and Strategy games. So if you are going to argue based on popularity alone, then HardOCP reviews miss that mark by miles.

You just stated yourself people don't play those old games anymore, why is it relevant to include the performance in those games as a buying decision? Not playing it, why do we care?

The games [H] benches are new AAA titles. I think it matters more than Unigine, 3dMark... and if you're going to discard these brand new games because a previous gen card can max them out, why you bother with dirt 3 and skyrim?

 

[RussianSensation]

why you bother with dirt 3 and skyrim?

GTX680 won in BF3, Dirt 3 and SKYRIM from PCPerspective review. I can't talk about games they didn't' test. The main point is it's a 2nd review that reinforces the same thing that Tom showed - GTX680 wins almost all the benchmarks so far. That's the key takeaway.

Yet you just declared the GTX 680 clearly superior by referencing 3DMark11, and two console ports, Dirt 3 and Skyrim... interesting.

3D Mark 11 is actually very good for predicting performance delta on average unlike Vantage was. See HD6970 vs. HD7970.

I used Dirt 3 and SKYRIM because those were the ones linked from PC Perspective. In SKYRIM, at the highest setting HD7970 was dipping in 40s vs. mid-60s for GTX680. That's a noticeable difference imho. For Dirt 3, GTX680 doesn't provide any more playability over HD7970. The advantage is simply academic.

Thanks for ignoring that GTX680 also costs $50 less and has more features + transferable lifetime warranty with EVGA though and missing that part of my post conveniently.

 

[Zebo]

So Intel should disable TurboBoost on all of its i5/i7 CPUs when reviewers test them against Bulldozer and when i3s are included in revies? This is literally free performance for everyone and you are against it? Grasping for straws much?

The reviews focus on what the consumer will get out of the box with both cards.

Look, it's been 4 consecutive generations of you owning an AMD card. Before one could make a very strong argument that AMD had better performance/$ with HD4870/4890/5850/5870/6950/6970 and power consumption was lower as a bonus. Not this time. This time GTX680 finally beats HD7970 in all metrics imaginable from trivial engineering metrics of performance/watt, performance/mm^2, performance/transistor to gaming focused metrics such as absolute performance, price, and features and it has lower power consumption. Plus EVGA offers lifetime transferable warranty, there is active Vsync, TXAA, 3D surround, PhysX, CUDA, faster video transcoding for iPad. All that for $50 less. HD7970 is good card, but it's time to move on.

If you had some bad experiences with NV cards (perhaps some bumpgate with GeForce 8/9), or just want to cheer for the underdog, then just say so.

GTX680 is clearly superior to HD7970 for gamers. This time it has HD7970 beaten in pretty much everything.

Well said. In the last 5 years I've never thought nV had as much lead as today even with superior performance at times with cards like GTX 580. Too much overhead to run them and dint have a good p/p ratio were the negatives. There are no negatives here from a average gamers prospective. I understand some compute stuff is weak but whatever maybe 1% care about that. These kepler cards and derivatives therof to come will be the best card from nV since 8800 series.

 

[Aeiou]

What's the ETA on reviews from all the big sites?

 

[Grooveriding]

I tend to prefer mostly just seeing the latest games, but I can see the value of some older games depending on what card you are reviewing. I really don't like to see sites that do an overall performance conclusion that review tons of games, many of which are dog awful old, because it skews the overall perf. conclusion.

If a GTX 580 is running freaking Wolfenstein at 200fps and a 6970 is running it at 130fps, I don't care because neither would have any benefit over the other for that game. But if you are throwing those results into an overall perf. summary number where it is skewing the results in favour of one card over the other, you've just rendered that overall summary absolutely worthless.

I mean if you insist on benching ancient games on the latest hardware where they are getting FPS in the 100s, at least do not create an overall perf summary using those numbers. If you must create such a summary, break it down into DX9, DX10 and DX11 games at the minimum, so you don't give what amounts to a final verdict using data that is worthless for giving a fair assessment in terms of the experience a buyer will have with one card or the other.

 

[RussianSensation]

I mean if you insist on benching ancient games on the latest hardware where they getting FPS in the 100s, at least do not create an overall perf summary using those numbers. If you must break it down into DX9, DX10 and DX11 games at the minimum, so you don't give what amounts to a final verdict using data that is worthless for giving a fair assessment in terms of the experience a buyer will have with one card or the other.

Ya, but see the irony here is the older games (Witcher 2, Crysis 1/Warhead, Crysis 2, Metro 2033, Dragon Age 2) are the more demanding games. So by including stuff like Mass Effect 3 and Deus Ex is actually doing exactly what you and I don't want to see - 100+ fps scores. The same thing as AT using Portal 2 in their review. Because HardOCP only tests 5 games, now 2/5th of their final scores is weighted towards their conclusion. If a site uses 14-15 games, then including 2 console ports only skews the average by 2/15th. So in fact, HardOCP's conclusions are FAR more biased statistically.

For example Computerbase will cover all the games tested in HardOCP's review + a bunch of other demanding games in different genres too.

Their reviews are well rounded and paint a better picture using a greater # of demanding games.

So my point is there is no problem using HardOCP as a data point, but it should never be used to make overall GPU recommendations since 5 games (2 of which are console ports) are not large enough of a sample size.

 

[blastingcap]

I can't remember which site does this, but they weight results so that no matter what you score, if it's below their "playable" line, you get no points. And if you get more than their upper limit, your advantage gets halved. So the stuff between 30 (I think? I can't remember) and 60 fps gets extra weight. It's a crude but somewhat effective method of getting rid of outliers and of penalizing cards that can't even get a playable result. I think they also have fps vs. time graphs too. Frankly I can do without everything but the time graphs.

 

[Ieat]

Well said. In the last 5 years I've never thought nV had as much lead as today even with superior performance at times with cards like GTX 580. Too much overhead to run them and dint have a good p/p ratio were the negatives. There are no negatives here from a average gamers prospective. I understand some compute stuff is weak but whatever maybe 1% care about that. These kepler cards and derivatives therof to come will be the best card from nV since 8800 series.

Yeah I remember browsing through this forum in the Summer/Fall of 2010 after the release of fermi and in anticipation of the ati 6 series. Comments such as "How is nvidia going to stay in business?" or "2 generations behind they will never catch up" were common from some. Fast forward a year and a half and things are looking a lot different.

 

[MrK6]

I tend to prefer mostly just seeing the latest games, but I can see the value of some older games depending on what card you are reviewing. I really don't like to see sites that do an overall performance conclusion that review tons of games, many of which are dog awful old, because it skews the overall perf. conclusion.

If a GTX 580 is running freaking Wolfenstein at 200fps and a 6970 is running it at 130fps, I don't care because neither would have any benefit over the other for that game. But if you are throwing those results into an overall perf. summary number where it is skewing the results in favour of one card over the other, you've just rendered that overall summary absolutely worthless.

I mean if you insist on benching ancient games on the latest hardware where they are getting FPS in the 100s, at least do not create an overall perf summary using those numbers. If you must create such a summary, break it down into DX9, DX10 and DX11 games at the minimum, so you don't give what amounts to a final verdict using data that is worthless for giving a fair assessment in terms of the experience a buyer will have with one card or the other.

I can't remember which site does this, but they weight results so that no matter what you score, if it's below their "playable" line, you get no points. And if you get more than their upper limit, your advantage gets halved. So the stuff between 30 (I think? I can't remember) and 60 fps gets extra weight. It's a crude but somewhat effective method of getting rid of outliers and of penalizing cards that can't even get a playable result. I think they also have fps vs. time graphs too. Frankly I can do without everything but the time graphs.

Hit the nail on the head. I care about pertinent performance - with this card, what can you expect in your machine with the most recent games? That's why I value [H] reviews, they play the game, and then show you exactly how it played for them. That's about the best data possible to use in a generalization; everything else, from synthetic tests to canned benchmarks, just stray further away from that ideal.

 

[Grooveriding]

Ya, but see the irony here is the older games (Witcher 2, Crysis 1/Warhead, Crysis 2, Metro 2033, Dragon Age 2) are the more demanding games. So by including stuff like Mass Effect 3 and Deus Ex is actually doing exactly what you and I don't want to see - 100+ fps scores. The same thing as AT using Portal 2 in their review. Because HardOCP only tests 5 games, now 2/5th of their final scores is biased. If a site uses 14-15 games, than including 2 console parts only skews the average by 2/15th. So in fact, HardOCP's conclusions are FAR more biased statistically.

For example Computerbase will cover all the games tested in HardOCP's review + a bunch of other demanding games in different genres too.

Their reviews are well rounded and paint a better picture using a greater # of demanding games.

Yeah for sure, once I spent some time looking closely at computerbase's reviews I saw just how amazing they are. Heck, they even include short sound bites of each card shown in each review under load, so you can a little bit of an idea what the acoustics are like :thumbsup: I was looking over their reviews all the way back to 8800GTX and 3870 and they have been so consistent with the way they give results that you can see how much more powerful AMD/NV's current cards are over the cards back then. If interested @ 2560 4xAA the 7970 is 675% stronger than a 3870 and the GTX580 is 373% stronger than the 8800GTX. It was pretty cool to be able to see that laid out by going through their reviews.

With all the GPU power around in the more expensive cards these days there are not many games that are pushing it much. Crysis 1/2, BF3, Witcher 2, Metro, Arma 2, Skyrim actually is pretty punishing on ultra with mods - I certainly get dips and run out of VRAM as well to boot.

I like [h] because for me personally, he does so many reviews at my resolution So selfishly, for me, the reviews are really helpful. lol There are also only so many sites that ever do CF/SLI benches using 3 cards, which I've always liked and [h] is one of them. The only others that come to mind are guru3d and here at anand.

 

[RussianSensation]

I can't remember which site does this, but they weight results so that no matter what you score, if it's below their "playable" line, you get no points. And if you get more than their upper limit, your advantage gets halved. So the stuff between 30 (I think? I can't remember) and 60 fps gets extra weight. It's a crude but somewhat effective method of getting rid of outliers and of penalizing cards that can't even get a playable result. I think they also have fps vs. time graphs too. Frankly I can do without everything but the time graphs.

Hexus.net

 

[amenx]

Going to the semiaccurate forums on a day like this is like going to redstate.com after the election of Obama: you feel very dirty, but it's totally worth it.

Or AMDZone on a major Intel release. You see a bit of the same panties-in-a-bunch crowd over here at AT as well.

 

[gwolfman]

Check out my post here:
http://forums.anandtech.com/showthread.php?p=33177271