GA-X79-UD7 or UD5: Confirm x16-x16-x8 SLI?

[eduncan911]

Can someone confirm that the Gigabyte GA-X79-UD7 or GA-X79-UD5 supports the following link speeds with 3-way cards?

x16 - x16 - x8

Getting conflicting information from NewEgg (saying x16 - x8 -x16 -x8), which we know isn't possible on the X79 chipset.

Looking at manual:

http://download.gigabyte.us/FileList/Manual/mb_manual_ga-x79-ud7_e.pdf

It has a component map that clear shows at 1 slot dedicated at x16 at all times. The other x16 is "switchable". Down in the manual on page 10, it says:

The PCIEX8_2 slot shares bandwidth with the PCIEX16_2 slot. When the PCIEX8_2
slot is populated, the PCIEX16_2 slot will operate at up to x8 mode.

Ok, that makes sense. PCIEX8_2 is the "4th" slot FYI. So, when the 4th slot is populated, it sounds like it drops PCIEX16_2 (the 3rd slot in teh system) down to x8 for a total of x16-x8-x8-x8. That makes absolute perfect sense.


But behold, then I read this in the manual, just a couple of lines down on the same page:

Support for 4-Way/3-Way/2-Way AMD CrossFireX™/NVIDIA SLI technology
* The PCIEX16 slots operate at up to x8 mode when 4-Way/3-Way AMD
CrossFireX™/NVIDIA SLI is enabled.

Whaaa? Now we are at x8-x8-x8-x8 in 4-way, and 3-way modes?


Please, can someone with 3-way SLI or Crossfire confirm the link speeds on the UD7 or the UD5? I plan on only running 3-way, for now (see sig).

Thanks!

 

[BigTerminator]

You have to understand that you will not get any performance difference between x8 and x16 speeds even in 3 way SLI. People have put a 590 into an x8 slot and even x4 and the difference wasn't noticeable at all.

With that out of the way, the diagram shows x16 x8 x16 x8 but the x16 slots will seperate into x8 if the if the x8 slot is also occupied so with 4 graphics card it will run at x8 x8 x8 x8 and with 3 it will run x16 x8 x8. But again you will not have any performance lost due to the card running at x8. Gigabyte opted not to use the NF200 chip because they thought it would add too much latency and hurt performance.