OFFICIAL KEPLER "GTX680" Reviews

[blastingcap]

Could someone here confirm that a) you do not need a displayport adapter to hook up 3 monitors (just DVI or HDMI), and b) whether the GTX 680 DOES or DOES NOT screen tear when driving 3 monitors? I mean single-GPU systems only, not SLI. I am hearing that the HD 7xxx series still has the tearing problem but that the GTX 6xx series does not, but I'd like confirmation by someone who actually has the card, if possible.

This could impact my buying decisions down the line, even though I am currently waiting for 28nm prices to fall before diving in.

Thanks!

 

[blastingcap]

Is it just me or does anybody else smell a thread lock and vacations coming?

680 is not midrange. When was the last time you saw a next gen midrange card decimate the previous gen highend ?

Huh? That happens sometimes. The GTX 460 was a bit faster than the GTX 280 and on par with the GTX 285 and had more oc headroom, so oc vs oc the GTX 460 was faster than the GTX 285. Then there was the 8800 series, do I need to say more?

I can think of other examples. I think it's pretty clear from Anandtech's review and the lack of HPC stuff on the GtX 680 that it was based off the previous-gen midrange part (GTX 560).

But it's a matter of semantics anyway. What I care about is availability, price/perf, efficiency, and lack of obnoxious things like needing displayport adapters or screen tearing. Other things like adaptive Vsync, too. Don't care about physx/CUDA though others might.

 

[toyota]

Is it just me or does anybody else smell a thread lock and vacations coming?

680 is not midrange. When was the last time you saw a next gen midrange card decimate the previous gen highend ?

sure its being sold as high end but everything about the gpu including its very chip name of gk104 tells you that this was not originally designed to be their top gpu. and it does not decimate the previous high end at all. its the smallest performance increase I ever recall from a true next gen card. about the only reason it looks impressive is because AMD set the bar so low by releasing the 7970 at 50% higher cost than 6970 while only increasing performance by 41%. if AMD had released a faster next gen card at around the same price point as its previous generation then the gk104 would have been released as a gtx660.

 

[BoFox]

I know it won't change much but switching HEAVEN for Wolfenstein seems like a fair thing to do. There are always some games that are outliers, but people play those games too. Doing that would make those cards switch places.

You got that right! Excluding Heaven and including Wolfenstein (a "freak" outlier) actually makes HD 7970 perform 0.2% better than GTX 680 at 2560x1600 when both cards are overclocked to the max on air cooling.

From the charts:

http://alienbabeltech.com/main/?p=29157&page=4

2560x1600-only differences (max OC GTX 680 vs max OC HD 7970)

0.0821256039 (Heaven)
-0.0801393728
-0.7878787879 (Wolfenstein, extreme outlier by far)
-0.076
-0.0144927536
0.0260416667
0.205
0.085399449
0.1420959147
-0.0625
-0.1623036649
-0.0717488789
-0.0131578947
-0.0615384615
0.1151736746
0.0980392157
0.2052631579
0.0526315789
0.1018957346
0.1020710059
-0.0490307868
0.2068965517


0.0019928615 - average difference (which is 0.2%)

AVERAGE DIFFERENCE excluding HEAVEN is -0.2%, so you got that right, Lepton87.

However, after taking all of the reviews out there on the 'net into account, I would take Wolfenstein to be a "freak outlier" (as you can see in the list of numbers above) and still consider GTX 680 to beat HD 7970 solely at 2560x1600 on average. An example of how Geforce cards do well in OpenGL games (other than Wolfenstein) is this:

http://hothardware.com/Reviews/NVIDIA-GeForce-GTX-580-A-New-Flagship-Emerges/?page=5

Personally, I would like to stress that excluding Wolfenstein changes it to 4.0% in favor of GTX 680, just at 2560x1600 for max O/C shootout.

By the way, if this shows ABT's reviews to be unbiased, it would be nice if somebody could reverse the actions taken and suggest to AMD that ABT should no longer be blacklisted since ABT's review actually shows OC'ed HD 7970 to beat OC'ed GTX 680 at 2560x1600 in 21 games overall (still passing the test of integrity after being blacklisted).

What's your favorite review of GTX 680 so far?

 

[Lepton87]

By the way, if this shows ABT's reviews to be unbiased, it would be nice if somebody could reverse the actions taken and suggest to AMD that ABT should no longer be blacklisted since ABT's review actually shows OC'ed HD 7970 to beat OC'ed GTX 680 at 2560x1600 in 21 games overall (still passing the test of integrity after being blacklisted).

What's your favorite review of GTX 680 so far?

I wouldn't call 0.2% advantage beating, it's within margin of error. Those cards at those clocks are perfectly tied. It's amazing how close those cards perform with such divergent architectures.

 

[aaksheytalwar]

IMO anything less than 5% is within a margin of error. Differences should count only if they are at least 5% or more.

Less than 5% isn't noticeable, period. And when you change the games it may change. Overclock3D did a review where a 7970 OC beat the Oced 680 by about 4%. Now this changes in this review to the exact opposite. So depending on the games used, the 680 may win or the 7970 may win.

At stock the 680 will win most. But when overclocked high enough the 7970 will win nearly equal.

Also, your 3D Mark scores for the 7970 are low. My 7970 at 1175 1625 does 9600+. So for a 1200 1575 the score should be at least 9600+ as well.

Anyway, anything less than a 15% definite improvement across the board is a side grade. It is possible to nit pick benchies such that the 7970 wins all and gains lead by 10-15% and the vice versa is also true. But when you compare say 50 or 100 games out of the all released over the last 1-2 years, they will probably diff by a median of < 10% and that too game dependent where each will have its share of wins. So even tho the 680 may be slightly better for benchmarking purposes, it doesn't make the gaming experience any different based on the FPS you get.

I agree 7970 to 685 is an upgrade. But 7970 ~ 680 or 680 ~ 7970 are both side grades once they are overclocked to about the same clock

I got the 3D Marks score difference, they used a 920

 

[BoFox]

IMO anything less than 5% is within a margin of error. Differences should count only if they are at least 5% or more.

Less than 5% isn't noticeable, period. And when you change the games it may change. Overclock3D did a review where a 7970 OC beat the Oced 680 by about 4%. Now this changes in this review to the exact opposite. So depending on the games used, the 680 may win or the 7970 may win.

At stock the 680 will win most. But when overclocked high enough the 7970 will win nearly equal.

Also, your 3D Mark scores for the 7970 are low. My 7970 at 1175 1625 does 9600+. So for a 1200 1575 the score should be at least 9600+ as well.

Anyway, anything less than a 15% definite improvement across the board is a side grade. It is possible to nit pick benchies such that the 7970 wins all and gains lead by 10-15% and the vice versa is also true. But when you compare say 50 or 100 games out of the all released over the last 1-2 years, they will probably diff by a median of < 10% and that too game dependent where each will have its share of wins. So even tho the 680 may be slightly better for benchmarking purposes, it doesn't make the gaming experience any different based on the FPS you get.

I agree 7970 to 685 is an upgrade. But 7970 ~ 680 or 680 ~ 7970 are both side grades once they are overclocked to about the same clock

I got the 3D Marks score difference, they used a 920

It's not like as if he used a 920 for the 7970 and then Sandy Bridge for GTX 680. GTX 680 would still be faster than 7970 in 3DMark on your rig. :\

 

[notty22]

It's not like as if he used a 920 for the 7970 and then Sandy Bridge for GTX 680. GTX 680 would still be faster than 7970 in 3DMark on your rig. :\

It's a combination of finding a needle in a haystack and a strawman comment

 

[BoFox]

It's a combination of finding a needle in a haystack and a strawman comment

Except that it was like a gigantic needle sticking out of the .. , in that HD 7970 scoring low in 3DMark was the "climax" of the post. It was to clarify for the thread's sake which only took 1 second to do (faster than finding the strawman picture).

Overclock3D's review is a needle in the haystack of reviews out there, and is within the 5% margin of error. Alienbabeltech runs each benchmark 3 times before deriving an average for each game bench to make sure the margin of error is minimized.

Do you guys want me to show a calculated average of all the differences of every single GTX 680 OC review out there? It's a service I'm willing to provide for those who are rather concerned.

 

[Tempered81]

Alienbabeltech runs each benchmark 3 times before deriving an average for each game bench to make sure the margin of error is minimized.

This is an off-kilter comparison though if he clocks the 680's memory at 7200mhz and the 7970's memory at 6200mhz. A Ghz difference, and they're still equal within 0.02%. CCC max is 1575, and that's where ABT stopped their overclock. Tell Mark to add some mvddc, Bo Fox.

 

[WhoBeDaPlaya]

exactly. AMD charged 50% more for 41% improvement over 6970. at least Nvidia gave you a 36% improvement while keeping same price as gtx580.

Do you really begrudge AMD for obliging its shareholders and maximizing profit while the competition was down?

We all knew that in all probability the 79x0 was due for a price adjustment once Kepler came out and it was a good deal (relative to the 580) _at the time_. You don't complain about how Micropolis was price gouging us on their new 100% larger drive and compare it to how Seagate is charging us less for their 100% larger drive do you?

 

[toyota]

Do you really begrudge AMD for obliging its shareholders and maximizing profit while the competition was down?

We all knew that in all probability the 79x0 was due for a price adjustment once Kepler came out and it was a good deal (relative to the 580) _at the time_. You don't complain about how Micropolis was price gouging us on their new 100% larger drive and compare it to how Seagate is charging us less for their 100% larger drive do you?

my concern is from the consumers point of view. all AMD has done with the 7850/7870 and 7950/7970 gpus is raise prices at the same rate as raising performance. that sucks for consumers because we usually get much faster cards at about the same price points as the previous gen.

 

[WhoBeDaPlaya]

my concern is from the consumers point of view. all AMD has done with the 7850/7870 and 7950/7970 gpus is raise prices at the same rate as raising performance. that sucks for consumers because we usually get much faster cards at about the same price points as the previous gen.

Fair enough and personally, that is the reason why the 79x0 wasn't enticing to me (already have enough performance with an unlocked and OCed reference 6950).

Drop the price to the $400s, and we got a deal

 

[railven]

my concern is from the consumers point of view. all AMD has done with the 7850/7870 and 7950/7970 gpus is raise prices at the same rate as raising performance. that sucks for consumers because we usually get much faster cards at about the same price points as the previous gen.

See, I still contest this thought pattern because everyone keeps ignoring that the GTX 580 didn't see price drops it's whole life on market - this is not normal.

I've asked everyone and yet no one has answered. Where would you price the HD 7970? At launch (or announcement) it beat the GTX 580 in all benches. Did you expect them to sell it for less (and yes I realize that nVidia did just that with the GTX 680, but if I were to listen to the other half of people, it's the GF114 replacement so it's literally the same thing AMD did.)

 

[BoFox]

This is an off-kilter comparison though if he clocks the 680's memory at 7200mhz and the 7970's memory at 6200mhz. A Ghz difference, and they're still equal within 0.02%. CCC max is 1575, and that's where ABT stopped their overclock. Tell Mark to add some mvddc, Bo Fox.

Apparently, he read this post, and replied to you:
http://alienbabeltech.com/abt/viewtopic.php?f=6&t=24550&view=unread#p62864

Also, when benching that many games, usually, the max overclock is a bit lower. For those who only bench 3-5 games, it's very easy to do it with unreasonably high clocks that's completely impractical for 8-hour gaming sessions across a broad spectrum of games (some being much more sensitive to overclocking).

 

[fstime]

New version of Precision X is out.

http://www.evga.com/precision/

 

[HurleyBird]

Personally, I would like to stress that excluding Wolfenstein changes it to 4.0% in favor of GTX 680, just at 2560x1600 for max O/C shootout.

Except that many 7970's go up to 1300MHz. 1200MHz max OC with voltage modding is at the very bottom of what you can expect. ABT's sample is below average quality.

 

[blackened23]

New version of Precision X is out.

http://www.evga.com/precision/

Still not sure why people use this over MSI afterburner (which works with the 680), which is better all the way around (with the better OSD and monitoring, better at everything pretty much).

It made sense when it was the only game in town, but why.

 

[BoFox]

Except that many 7970's go up to 1300MHz. 1200MHz max OC with voltage modding is at the very bottom of what you can expect. ABT's sample is below average quality.

You would be hard-pressed to find a 7970 review that shows it overclocked to 1300MHz that is tested to be stable in more than 5-7 games. On the forums, most of these 7970's are water-cooled for 24/7 stable gameplay at 1300MHz.

ABT's sample comes from Newegg, and not directly from AMD. Moreover, I would still expect it to be a fairly average representative of a retail HD 7970 (not a cherry-picked one sent to review sites) that is actually tested in 21 games and 3 synthetic benchmarks, run 3 times each in 2 different resolutions. Avid overclockers without an agenda will admit that for stable 24/7 game play with several games requires about 5% (edit, or even 15-20% as with my experience) lower overclock than simply testing with Furmark or Unigine for 1 hour, let alone running a quick game benchmark for a few minutes like many review sites out there do.