Bus Speeds and PCI Lanes

[the1corrupted]

So today I'd like to start a discussion on the principle of bus speeds in correlation to the number of PCI lanes available.

Say you have two graphics cards that are identical in every way except for the bit rate on the bus. (Ie, 660 Ti and 670). Without any regard given to price, general performance, overall compatibility with games, and so on... The topic is this:

With a 192-bit bus interface, sure the card would be able to process less than a 256 bit utilizing all 16 lanes of PCI express operation, but here's the catch. If we put two of each card into an SLI capable board that reduces their operation to 8x8 (which would be the most common/accessible SLI), would the performance be exactly identical? I have seen benchmarks where the 660 Ti and the 670 meet each other within the acceptable 'margin of error' of 1% or 3% where they were supported.

 

[lt_dan_zsu]

No, there aren't any bottlenecks at this point with x8 3.0 lanes. So a 670 in an x8 lane would get this same performance as in x16. I have seen benchmarks doing this with a 680. The difference with the 680 in a 3.0x16 lane vs a 2.0x8 lane was something like 1 frame per second. Which could easily just be within the margin of error for benchmark programs.
http://www.techpowerup.com/reviews/Intel/Ivy_Bridge_PCI-Express_Scaling/6.html
That review shows that there is only in 10% bottleneck with a 680 in a pcie 1.1x4 slot vs 3.0x16. In all honesty pcie 3.0 really won't be necessary for new cards for a few years probably 560ti's don't even get bottlenecked in 2.0x1 slots! The reason the performance was similar was because the 7870, 660ti, 7950, 670, 680, and 7970 aren't really that far apart from each other.

 

[the1corrupted]

No, there aren't any bottlenecks at this point with x8 3.0 lanes. So a 670 in an x8 lane would get this same performance as in x16. I have seen benchmarks doing this with a 680. The difference with the 680 in a 3.0x16 lane vs a 2.0x8 lane was something like 1 frame per second. Which could easily just be within the margin of error for benchmark programs.
http://www.techpowerup.com/reviews/Intel/Ivy_Bridge_PCI-Express_Scaling/6.html
That review shows that there is only in 10% bottleneck with a 680 in a pcie 1.1x4 slot vs 3.0x16. In all honesty pcie 3.0 really won't be necessary for new cards for a few years probably 560ti's don't even get bottlenecked in 2.0x1 slots! The reason the performance was similar was because the 7870, 660ti, 7950, 670, 680, and 7970 aren't really that far apart from each other.

Thanks for the info! 😀