ASUS GeForce GTX 680 SLI

[N3M3515]

It needs to be there for consistency with the all-resolution/average Relative Performance figure.

I think that it is irrelevant..
For the average only should be 1680x1050 - 1920x1080 - 2536x1xxx
nobody buys a GTX680 to play at 1024x768 or 1280x1024

 

[Kreij]

It gives you an average based on all of the various aspect ratios (4:3, 16:10, etc.)
Whether 1024x768 is a good 4:3 to use, is not for me to judge.

 

[mR Yellow]

Another incomplete review... where is the 7970 CFX?
Other sites show it being faster than SLI.

 

[N3M3515]

W1zz, it wouldn't hurt to put 7970 & 7950 cfx results, for the sake of comparison.

 

[m1dg3t]

Patience all, i'm sure they are coming

 

[MxPhenom 216]

Another great review W1zz but first paragraph:

"NVIDIA launched its newest, most advanced GPU, the GeForce GTX 680, on March 22. The card wiped the floor with other single-GPU cards, but there's a lot more to it than its performance"

Wiped the floor? Really? Faster than 7970 yes in most case's but by very slight margin's

god you don't need to get so specific. Its performance FPS was wasn;t that much but when you consider other things like noise, power consumption, cooling, and for the price it wiped the floor pretty good

If wizz says a part wiped the floor, it wiped the floor!

 

[erocker]

Defend the billion dollar corporation, DEFEND IT! Lol, I want to see how a 4gb version fares in higher resolution gaming.

 

[MxPhenom 216]

Defend the billion dollar corporation, DEFEND IT! Lol, I want to see how a 4gb version fares in higher resolution gaming.

well the 2gb version is holding its own pretty well, so if thats any indication on the 4gb, i assume itll do a bit better

 

[aBigRat]

It seems, when comparing clock to clock. The GTX 680 is at HD 7950 level.
800MHz vs 1050MHz+ (25% clock gap) and 60% vs 71% (22% performance gap).
And the real transistor count of the two is almost equal, and so does the power draw.
Hope AMD will release some variant like HD 7960 and 7980 with rated clock at 1GHz.
That will help AMD retake the crown nicely.

 

[HTC]

W1zz, it wouldn't hurt to put 7970 & 7950 cfx results, for the sake of comparison.

As well as 6970s and 580s dual cards.

 

[m1dg3t]

god you don't need to get so specific. Its performance FPS was wasn;t that much but when you consider other things like noise, power consumption, cooling, and for the price it wiped the floor pretty good

If wizz says a part wiped the floor, it wiped the floor!

If you are happier than anything to have a stripped down, limited, OC'd Fermi that barely perform's better, MOAR power to ya 15% - 20% is starting to wipe the floor, like the diff between 7970 & 680 vs 580 & 6970 And i'm not just refering to FPS

I'm still waiting for Kepler

See: I can troll too

 

[erocker]

Trolling is lame, I like graphs and charts and stuffs. BF3:

If the GTX 680 does this without hitting this amout of vram, I'm sure Nvidia did some "magical" IQ tweaking... Unless the game engine itself scales IQ due to Vram limitations.. Don't know really.

 

[HTC]

Trolling is lame, I like graphs and charts and stuffs. BF3:

http://i403.photobucket.com/albums/pp112/erocker414/bf3vramusage.jpg

If the GTX 680 does this without hitting this amout of vram, I'm sure Nvidia did some "magical" IQ tweaking... Unless the game engine itself scales IQ due to Vram limitations.. Don't know really.

I said it before in post #29:

It surprised me that, as the resolution went up, so did the scaling. Since it had 2 GB (as opposed to 7970's 3GB), i was expecting it the other way around.

 

[erocker]

It surprised me that, as the resolution went up, so did the scaling. Since it had 2 GB (as opposed to 7970's 3GB), i was expecting it the other way around.

I don't understand what you're saying here. Are you agreeing that IQ goes down with the memory limitation?

 

[HTC]

I don't understand what you're saying here. Are you agreeing that IQ goes down with the memory limitation?

I'm agreeing that, since the scaling gets better with higher resolutions, somethings different between the 2, unless nVidia accomplishes the same with less RAM (doubtful).

EDIT

I'm surprised nobody's investigating how's it possible to scale better @ higher resolutions with less RAM.

Maybe the IQ gets lowered, maybe nVidia has a better implementation of it's scaling then ATI does, maybe something else is going on: dunno, really.

 

[mandis]

What was the point of this review? Why would you compare an SLI configuration solely against single cards? What is that supposed to prove? How are we supposed to gain any tangible evidence from this??

I was looking forward to this review and now I can't justify the time i wasted reading it. I'm very dissapointed!

 

[BumbleBee]

the natives are getting restless

 

[Stevethegreat]

With all due respect I just don't think that the scaling is ought to be so slight. Excluding driver's problems you should be getting upwards the 70% compared to single card performance. It seems to me that you're bottlenecked by the CPU. Maybe it's time to retire the i7-920. The good ol' chap has seen too many battles (reviews) and maybe lately he's fudging your results.

Try a Sandy Bridge at 4.7GHz (or even better an Ivy Bridge) and you'll get different results me thinks, especially on lower resolutions. Let those little beasts -your GFX cards- breathe.

Apart from that, great review as always very methodic and the most informative compared to all the other sites...

 

[jaredpace]

What was the point of this review? Why would you compare an SLI configuration solely against single cards? What is that supposed to prove? How are we supposed to gain any tangible evidence from this??

I was looking forward to this review and now I can't justify the time i wasted reading it. I'm very dissapointed!

Just check both reviews and make your own comparison like this:

 

[mandis]

Just check both reviews and make your own comparison like this:

http://tpucdn.com/reviews/ASUS/GeForce_GTX_680_SLI/images/crysis_2560_1600.gif
http://tpucdn.com/reviews/ASUS/HD_7970_CrossFire/images/crysis_2560_1600.gif

Thanx jaredpace!

I suppose what bothered me the most with this review was its "MARKETING" over "informative" orientation. This review should have been about last gen vs current gen SLIs as well as SLI vs CF. Anything less than that renders the entire attempt rather pointless...

 

[N3M3515]

With all due respect I just don't think that the scaling is ought to be so slight. Excluding driver's problems you should be getting upwards the 70% compared to single card performance. It seems to me that you're bottlenecked by the CPU. Maybe it's time to retire the i7-920. The good ol' chap has seen too many battles (reviews) and maybe lately he's fudging your results.

Try a Sandy Bridge at 4.7GHz (or even better an Ivy Bridge) and you'll get different results me thinks, especially on lower resolutions. Let those little beasts -your GFX cards- breathe.

Apart from that, great review as always very methodic and the most informative compared to all the other sites...

I agree in the part that sli and cfx will be bottlenecked with the i7 920. at least in 1080p

 

[Moatsim]

Omg

 

[xorbe]

is there a point testing 1024x768 lol

Though the result by itself isn't so interesting, it's very nice info to see how the gfx card perf scales 1024 through 2560

 

[xenocide]

I agree in the part that sli and cfx will be bottlenecked with the i7 920. at least in 1080p

The 920 is overclocked to 3.8GHz for all the tests, I have severe doubts it bottlenecks anything. If he went to SB people would complain since it's not 16x\16x. People are always just looking for any reason to refute the facts. The fact remains, W1z's test bench is fine, and his testing methodology is definitely amongst the best.

 

[N3M3515]

The 920 is overclocked to 3.8GHz for all the tests, I have severe doubts it bottlenecks anything. If he went to SB people would complain since it's not 16x\16x. People are always just looking for any reason to refute the facts. The fact remains, W1z's test bench is fine, and his testing methodology is definitely amongst the best.

I'm not refuting anything , nor looking for any reason to refute any fact.

Yes, W1z methodology is fine. But not perfect, not saying it has to be, but there is ALWAYS room for improvement, and i believe everyone is entitled to have its own opinion on any given topic.

So, my opinion is that i7 920 even at 3.8, could be a bottleneck at 1080p when testing highend single gpu cards in sli or cfx, even more if it is dual gpu cards in sli/cfx.