SLI & CFX PCIe Bandwidth Perf. - x16/x16 vs. x16/x8

smackababy

Lifer
Oct 30, 2008
27,024
79
86
So, basically 480GTX SLI isn't saturating the x16/x8 lanes? I know there was a 5870 CFX review awhile back that did show a small hit in FPS, but this one isn't showing much.
 

Termie

Diamond Member
Aug 17, 2005
7,949
48
91
www.techbuyersguru.com
I was under the impressing at least CFX ran at 8x/8x if one of the slots were 8x. This might just be for P55, but I thought it was for all.

This is exactly what confused me about this article. Maybe you get x16/x8 if you are using the third slot in an x58 board for some odd reason (spacing, for example), otherwise, I don't know that you'd ever get x16/x8. I think their x8/x8 article will be more relevant to most people (including me). Then again, I wouldn't be surprised if the results are exactly the same, because I'd think that both cards would be limited to x8 performance if one were limited by x8.
 

Elganja

Platinum Member
May 21, 2007
2,143
24
81
I run mine in the 1st and 3rd slot... it says 16x @ 16x for #1 and 16x @ 8x for the second one... however I was also under the impression that the slowest speed would be the max speed, so 8x
 

NoQuarter

Golden Member
Jan 1, 2001
1,006
0
76
The PCI-e bus only reaches threshold during large texture transfers when the VRAM doesn't happen to have the textures already loaded. During most gameplay the actually bandwidth requirements aren't that great, so what it helps most with is minimum frame rates when textures are loaded on the fly.
 

busydude

Diamond Member
Feb 5, 2010
8,793
5
76
If the current cards(SSD, GPU's) cannot saturate PCIe bus, then what is the need for the next gen bus i.e, PCIe 3.0?

I might be wrong, but isn't PCIe 3.0 being used in the upcoming chipsets for their SB processors?
 

Termie

Diamond Member
Aug 17, 2005
7,949
48
91
www.techbuyersguru.com
Update: HardOCP has posted a new article that compares x16/x16 to x8/x8 using GTX480SLI.

Verdict: it only makes a difference at triple-monitor resolutions. The other interesting finding: x8/x8 is actually slower than x16/x8. I would have thought that an x16/x8 setup would default down to x8/x8 performance, but apparently it doesn't.

Edit: see link below.
 
Last edited: