• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

GTX 680 2x to 3x SLI Scaling

H

heatlesssun

Extremely [H]
Joined
Nov 5, 2005
Messages
44,154
680%202x%20to%203x%20SLI%20Scaling.jpg


Crysis 2 2D & 3D
Ultra in game settings
4X AA 16 AF set in nVidia Control Panel

Battlefield 3 2D
Max settings 4x AA in game

Battlefield 3 3D
Max settings No AA

Batman Arkham City 2D
DX 11 Features MVSS & HBAO & High Tessellation
Detail Level Extreme
PhysX Normal

Batman Arkham City 3D
DX 11 Features Off
Detail Level Extreme
PhysX Normal

Nothing too rigorrous here, just ran through about 5 minutes in the same section in every game a few times and took one set readings as they were all pretty much consistent. The exception was Batman AC in 3D. Not sure what happened there but it looked like I was getting negative scaling there. So I replayed a different section there in the 3x SLI test but did it for ten minutes and got into more fighting. So the Batman 3D numbers are the iffiest and it was the one game that I didn't feel any noticeable improvement even though there was some scaling.

I'm not trying to sell the notion of 680 3x SLI. The scaling doesn't seem to be great on x58 but there is enough to make BF 3 and Crysis 2 noticeably more playable at some pretty high settings. 4x MSAA in BF 3 in 2D really wasn't playable though in 2 or 3 way. I didn't bench it but 2x MSAA did feel more playable in 3 way.

I would suspect that if you really want to go for 3x or 4x SLI you would probably see better results on an X79 platform. But at the same time Heaven 3 scaled at nearly 100% efficiency, so the potential is there and my guess is that drivers will improve this as well. But still, just some amazing performance and game play is very smooth, no microstutter and only the occasional hit or slow down with a lot of action like lot of explosions in BF 3.
 
Broken images...can't see any results? Going quad SLI 680s this week, myself. :)
 
Sweet on the quad SLI! The image is on a public Skydrive folder, I've tried it from machines where I wasn't logged into to and the link is working.
 
heatlesssun said:
680%202x%20to%203x%20SLI%20Scaling.jpg


Crysis 2 2D & 3D
Ultra in game settings
4X AA 16 AF set in nVidia Control Panel

Battlefield 3 2D
Max settings 4x AA in game

Battlefield 3 3D
Max settings No AA

Batman Arkham City 2D
DX 11 Features MVSS & HBAO & High Tessellation
Detail Level Extreme
PhysX Normal

Batman Arkham City 3D
DX 11 Features Off
Detail Level Extreme
PhysX Normal

Nothing too rigorrous here, just ran through about 5 minutes in the same section in every game a few times and took one set readings as they were all pretty much consistent. The exception was Batman AC in 3D. Not sure what happened there but it looked like I was getting negative scaling there. So I replayed a different section there in the 3x SLI test but did it for ten minutes and got into more fighting. So the Batman 3D numbers are the iffiest and it was the one game that I didn't feel any noticeable improvement even though there was some scaling.

I'm not trying to sell the notion of 680 3x SLI. The scaling doesn't seem to be great on x58 but there is enough to make BF 3 and Crysis 2 noticeably more playable at some pretty high settings. 4x MSAA in BF 3 in 2D really wasn't playable though in 2 or 3 way. I didn't bench it but 2x MSAA did feel more playable in 3 way.

I would suspect that if you really want to go for 3x or 4x SLI you would probably see better results on an X79 platform. But at the same time Heaven 3 scaled at nearly 100% efficiency, so the potential is there and my guess is that drivers will improve this as well. But still, just some amazing performance and game play is very smooth, no microstutter and only the occasional hit or slow down with a lot of action like lot of explosions in BF 3.
Click to expand...

Its possible you were CPU bottlenecking the 3 680 gtx's. Although heaven 3.0 getting near 100% scaling would negate that theory.
 
Image is working for me now. Strange.

I don't think it could be a CPU bottleneck issue. The i7-980x is a hexa-core that is typically only 10-20% slower than the (now) top-of-the-line i7-3960x. Although perhaps heatlesssun is right, X79 would help for the additional benefits with PCIe 3.0, etc.

I play BF3 with FRAPS on and just from my eyeballing and experience in-game, I'd say it looks like we are playing at about the same performance level. Same resolution - 5760x1080...and I too had to use 2x MSAA.

A better guess would be early driver issues. Nvidia will get performance fixed for tri/quad SLI in future release. My guess is their focus was on single card and SLI performance initially.

Good stuff! Thanks for the post.
 
sk3tch said:
Image is working for me now. Strange.

I don't think it could be a CPU bottleneck issue. The i7-980x is a hexa-core that is typically only 10-20% slower than the (now) top-of-the-line i7-3960x. Although perhaps heatlesssun is right, X79 would help for the additional benefits with PCIe 3.0, etc.

I play BF3 with FRAPS on and just from my eyeballing and experience in-game, I'd say it looks like we are playing at about the same performance level. Same resolution - 5760x1080...and I too had to use 2x MSAA.

Good stuff! Thanks for the post.
Click to expand...

Well the thing is, he is only clocked at 4.3ghz. Most games dont even take adavatange of the 2 extra cores.

And the I7-3960x can easily hit 5ghz. So being 10-20% slower clock for clock, and say being 700mhz lower It is possible to have a CPU bottleneck.

I dont think pci-e 3.0 > 2.0 will make any difference at all.
 
DASHlT said:
I dont think pci-e 3.0 > 2.0 will make any difference at all.
Click to expand...

It's not necessarily PCIe 3.0 speed-wise it's about the additional lanes of PCIe being available for multi-GPU implementations in X79.
 
sk3tch said:
It's not necessarily PCIe 3.0 speed-wise it's about the additional lanes of PCIe being available for multi-GPU implementations in X79.
Click to expand...

Well the X58 chipset can easily do 3 16x PCI-E slots. So I still dont see a problems with lanes
 
Seems odd that the minimum framerates for the 3-way versus 2-way are sometimes significantly lower.
 
Thanks for the info heatlesssun. Def some game driver work needs to be done imo, I would find it hard to believe a 4.3GHz cpu (whether it's quad or hex) or the X58 chipset would be the cause for this lack of game scaling, and the canned bench dispells that as DASHIT pointed out.
 
DASHlT said:
Well the X58 chipset can easily do 3 16x PCI-E slots. So I still dont see a problems with lanes
Click to expand...

x58 can only do that with a 3rd party pci-e controller, as x58 has only 40 lanes of pci-e 2.0. For 3 way sli, the board he used only allows pci-e 2.0 8x to the second and third card. So pci-e bandwidth may actually be a factor.
 
CrazyRob said:
x58 can only do that with a 3rd party pci-e controller, as x58 has only 40 lanes of pci-e 2.0. For 3 way sli, the board he used only allows pci-e 2.0 8x to the second and third card. So pci-e bandwidth may actually be a factor.
Click to expand...

I might stand corrected for the OPs board "3 x PCIe 2.0 x16 (at x16/x8/x8 or x16/x16/x1 mode)" for ASUS P6X58D Premium and if it's in that x1 mode, hell ya, I can understand these results. We certainly are finally seeing some math diff in PCIe these days and X58 is beginning to show its age. The canned bench scaling on this same system is still the elephant in the room though.
 
Quagmire LXIX said:
I might stand corrected for the OPs board "3 x PCIe 2.0 x16 (at x16/x8/x8 or x16/x16/x1 mode)" for ASUS P6X58D Premium and if it's in that x1 mode, hell ya, I can understand these results. We certainly are finally seeing some math diff in PCIe these days and X58 is beginning to show its age. The canned bench scaling on this same system is still the elephant in the room though.
Click to expand...

I don't have that particular board but on mine it was 16x/8x/8x by default. I switched it to 16x/16x/1x in the BIOS to get the most out of 2-way SLI.

I honestly don't think 8x vs 16x is going to make that big of a difference, either.
 
Agreed Maverick, don't see x16/x8/x8 causing it. Cool to have the option on the board, but since it's by default, I wonder how many ppl going SLi on your board never fiqured out/researched what you did and have been gaming x16/x8 SLi all along.

My board was straight forward and auto configured by how many cards are installed. His TRi is just crippled and a x16/x16/x1 would be an obvious enabler when looking at those results.

Still, that damn elephant?
 
yeah, it's hard to say. Historically it shouldn't make a huge difference, but being an entirely new architecture we won't know until tested. Unfortunately, 301.10 drivers disable pci-e 3.0 support on x79 platforms, so until that's fixed or we get our hands on some ivy bridge cpu's we don't have a way to test to make sure :(

*edit* not entirely true, boards with nf200 chips like the first gen evga x58 classified would provide true (although not native) 3x pci-e 2.0 x16. I haven't seen any tests on those, though.
 
Last edited:
Just to clear up the confusion the lanes are at 16x/8x/8x, just like they have been always as this is the 4th 3x SLI setup I've had on this board.
 
You must log in or register to reply here.
Back
Top