(hardware.info) Nvidia GeForce GTX 690 SLI review: ridiculously powerful

[Final8ty]

http://uk.hardware.info/reviews/2975/nvidia-geforce-gtx-690-sli-review-ridiculously-powerful

 

[Arkadrel]

And yet....

Battlefield 3 - Ultra 4xAA - DoF 90 1920x1080

3 x 7970 = 150 fps avg. (costs ~1200 $ (399x3))
4 x 680 =131 fps avg. (costs ~2000 $ (499x4))
3 x 680 = 128 fps avg. (costs ~1500 $ (499x3))
2 x 690 = 124 fps av.g (costs ~~2000 $ (999x2))

 

[BTRY B 529th FA BN]

Yeah, those numbers seem off

 

[BallaTheFeared]

And yet....

Battlefield 3 - Ultra 4xAA - DoF 90 1920x1080

3 x 7970 = 150 fps avg. (costs ~1200 $ (399x3))
4 x 680 =131 fps avg. (costs ~2000 $ (499x4))
3 x 680 = 128 fps avg. (costs ~1500 $ (499x3))
2 x 690 = 124 fps av.g (costs ~~2000 $ (999x2))

And yet...

Crysis 2 - Very High - Edge AA - 5760x1080


4x 7970 = 40.3 fps avg. (cost ~$1720 (430x4))

2x 670 = 52ish fps avg. (cost ~$750 (375x2))

 

[blackened23]

More proof that tri and quad scaling on both side is messed up a lot, preacher would not approve. Two GPUs is the sweet spot

 

[RussianSensation]

More proof that tri and quad scaling on both side is messed up a lot, preacher would not approve. Two GPUs is the sweet spot

Agree!

3-4 way CF and SLI scaling is terribad on both sides.

AvP
HD7970 x2 = 59.8
GTX680 x3 = 63.8

Batman AC Medium
HD7970 GE x1 = 56 :thumbsup:
GTX680 x4 = 62
HD7970 x4 = 43

BF3 Medium
7970 x2 = 84
GTX680 x4 = 93

BF3 Ultra
HD7970 x3 = 64
GTX680 x4 = 57

Crysis 3 Extreme
7970 x2 = 39
7970 x3 = 39
7970 x4 = 39
GTX680 x4 = 83 :thumbsup:

Hard Reset Ultra
7970 x3 = 75
690 x2 = 79
690 x1 = 89 :thumbsup:

Metro 2033 Medium
7970 x2 = 44
680 x3 = 48

Metro 2033 High :thumbsdown:
680 x2 = 21
680 x3 = 22
680 x4 = 23
690 x2 = 24
7970 x2 = 25
7970 x4 = 41

SKYRIM High - CPU bottleneck? D:
7970 x2 = 89
7970 x3 = 89
7970 x4 = 91
690 x1 = 91
680 x2 = 93
680 x3 = 96
but then....
680 x4 = 130.....

Shogun 2 Ultra
7970 x2 = 34
680 x2 = 42
690 x2 = 42
7970 x4 = 65

 

[Smoblikat]

Im not sure how this is news to us. Is it really that surprising that 4 graphics cards have some horsepower?

 

[BallaTheFeared]

I'd like to say it's surprising that 3/4 cards are doing so poorly for both sides, but it's really not, this generation has done nothing but disappoint hardware wise, why would the drivers be any different?

Last gen you'd see diminishing returns from AFR and bottlenecking, but you'd still typically see advantages from the 3rd and 4th cards, this generation having more than two cards actually hurts you more than it helps.

 

[Grooveriding]

Of course 2x 690 will be powerful. Scaling has always showed a significant drop off on the 3rd card though and a lot more when you add a 4th card. Don't think I will ever run 3 cards again, especially with the state of games. If you're running a resolution higher than 2560x1600P sure, but I find two 680s can run anything I want with any settings I want.

 

[RussianSensation]

If PS4/Xbox 720 pull off an HD6670 level GPU, then I bet by 2014, 2x $600 Maxwell cards may be enough to last the entire console generation. The game changer could be if 2160P screen prices drop to affordable levels. However, it may be 5+ before that happens since they cost about $30,000-40,000 today. Perhaps I must not understimate the power of programmers to cripple performance with next gen console ports.

 

[boxleitnerb]

I believe there are quite a few games where there is a CPU bottleneck and a VRAM bottleneck. It would be interesting to see what happens when you have 4GB cards.

Is there any info about drivers?

 

[moonbogg]

Two 690's are good for literally nothing. They don't fold, they don't game, they don't do crap but cost money. Not enough Vram for the theoretical GPU power anyway. If someone wanted 4 GPUs they'd be better off with 4gb 670's.

 

[Grooveriding]

If PS4/Xbox 720 pull off an HD6670 level GPU, then I bet by 2014, 2x $600 Maxwell cards may be enough to last the entire console generation. The game changer could be if 2160P screen prices drop to affordable levels. However, it may be 5+ before that happens since they cost about $30,000-40,000 today. Perhaps I must not understimate the power of programmers to cripple performance with next gen console ports.

That or widely available 120hz 1600P & 1440P panels from a reputable manufacturer. Even that will be a small niche as well though. The GTX 680 could be the way of the future. Small and power efficient GPU aiming for the perfect balance of performance/efficiency on the newest nodes and just throwing out the idea of a consumer GPU that is as powerful as possible because there is just no use for it anymore. Never mind that there would be no price changes either, still nvidia would charge $500 for it

Probably won't happen, but if games continue to stagnate..

 

[blackened23]

I'm about to give up on a 2560x1600 120hz panel, I want one. I love my existing 2560 screens though, I feel like panel makers just aren't motivated to innovate here (and add 120hz)

What excites me most are the 4k panels, hopefully those will see mass production next year. Viewsonic has a 4k model out now but it costs more than most cars lol.

 

[Final8ty]

I'd like to say it's surprising that 3/4 cards are doing so poorly for both sides, but it's really not, this generation has done nothing but disappoint hardware wise, why would the drivers be any different?

Last gen you'd see diminishing returns from AFR and bottlenecking, but you'd still typically see advantages from the 3rd and 4th cards, this generation having more than two cards actually hurts you more than it helps.

It is still to early to conclude that more than 2 with this gen in performance is a no go when AMD has only really just got it working with the 12.6 and 12.7 betas.
And look what they managed on single cards with performance jump with the latest drivers, i should not count my chickens yet on Multi gpu being the best as its going to get yet..

 

[Final8ty]

2560x1600 1x5970 Ares blue v 2x5970 Ares green 5 series.

 

[Final8ty]

 

[BallaTheFeared]

Exactly, drivers are currently trash on the newer cards, my 470s scale very well, but I have a good cpu driving them. Scaling is as awful as it ever was though once you get cpu bound.

The 7xxx series paper launched in December 2012, eight months later and in OLD titles they're still failing. Nvidia isn't much better if any, I don't even care to look it's just awful.

/Praise the establishment

 

[Final8ty]

Exactly, drivers are currently trash on the newer cards, my 470s scale very well, but I have a good cpu driving them. Scaling is as awful as it ever was though once you get cpu bound.

The 7xxx series paper launched in December 2012, eight months later and in OLD titles they're still failing. Nvidia isn't much better if any, I don't even care to look it's just awful.

/Praise the establishment

That's why i try not to jump on new gen, jump on the next gen that is a refresh with lessons learnt.

 

[Annisman*]

I myself was burnt pretty good by adding a third gtx 670 4gb, if some of you remember. Even at triple monitor resolution. the driver support just isn't there yet.

 

[T_Yamamoto]

"Ridiculous powerful"
It is pretty much 4 680s

 

[Grooveriding]

I myself was burnt pretty good by adding a third gtx 670 4gb, if some of you remember. Even at triple monitor resolution. the driver support just isn't there yet.

Kepler is much worse off than Fermi was for triple card scaling. This was apparent in early reviews and most assumed drivers would improve it, but they haven't. While triple/quad scaling has never been near as nice as dual is, Fermi did a good deal better than Kepler does.

 

[3DVagabond]

Last gen triple gpu seemed to be pretty good, scaling wise and, according to TechReport anyway, suffered less from microstutter. I think it's just that both Kepler and GCN are new and a fairly large departure from prior tech for both companies that it'll take a little while to mature.

Besides, a pair of Tahitis or GK104's running 1.2-1.3 GHz isn't good enough for you, then you're just a freak anyway.

 

[VirtualLarry]

how much of the lack of scaling above two gpus is due to pci-e bandwidth limitations?