thetacowarrior

Thanks for the rest of the QPI info as well as the vtt lesson, they are both very good pieces of info to know. So you think that 4.2 is high enough QPI for my 260 Core 216's in SLI?
I don't think its bottlenecking, I turned it up and benched, it didn't change more than the normal variance from run to run, but that's not an end-all answer to whether or not its bottlenecking, right?

QPI has loads of bandwidth, its really something that was designed with the server segment in mind since Intel merged the server segment with the high-end enthusiast segment with Nehalem/Bloomfield.  The normal QPI speed should be plenty for any desktop configuration even using 4 GTX 480's in 4-Way SLI.

4.2GT/s should be plenty for your 2 GTX 260 Core 216's in SLI. If you have the BClock overclocked them it will actually be running faster than 4.2GT/s since its really a multiplier setting more than an actual QPI setting, the reason its listed as 4.2GT/s is that 16x the defualt BClock of 133MHz's equals 4200 MT/s which is 4.2GT/s.  Its similar to the way the Memory Frequency is found and the CPU Frequency, you take the BClock and them multiply it by the multiplier thats set (you them double that speed to find the QPI speed as the QPI is bi-directional, it has 2 way traffic so can do 2 things per clock).

4.2 GT/s = 16x Bclock (4200/2 = 2100, 2100/133 = 16)
4.8 GT/s = 18x Bclock (4800/2 = 2400, 2400/133 = 18)
6.4 GT/s = 24x Bclock (6400/2 = 3200, 3200/133 = 24)

So for example if the BClock was set to 200 and you set QPI to 4.2 GT/s in the BIOS the actual QPI speed would be 16x200 = 3200, 3200x2 = 6400, the QPI speed would be 6.4 GT/s.

Your BClock if 153 and a setting of 4.2 GT/s for QPI would mean an actual QPI speed of 16x153 = 2448, 2448x2 = 4.9 GT/s.

what you dont want is for your QPI speed to be the same speed as the Uncore (NorthBridge, Memory Controller), if they are both at a 1:1 ratio (both at the same speed this can cause instability, i think its more of an issue on Bloomfield than Lynnfield since its easy to have the QPI and NB/IMC run at the same speed).  I mean the same clock speed and not the QPI's effective or data speed, a QPI data/effective speed of 6.4 GT/s is actually a QPI clock Speed of 3200MHz, if you had the Memory frewuency set to DDR3 1600 on the Bloomfield that would mean a NorthBridge/Integrated Memory Controller (IMC) frequency of 3200MHz (1600x2 = 3200), both the QPI clock frequency and IMC frequency would be running at 3200MHz which was rumoured to cause instability.