i am intrigued as well, i just cant find a solid source of info that explains just the NF200

How would it connect to the processor though, on a P55 platform without a northbridge and the PCH only offering 4x PCI-e 1.1?  Seems like a rather strange chip/setup to me.

from my review at Hardware Canucks (link)...



It seems kind of complicated, but to keep it simple, remember...only 16 lanes go to the CPU.

There is no way to make more than 16 lanes go to the CPU.

So what happens is with a single card in the top slot, that slot runs at 16X so all 16 lanes are being used by it. In the diagram above you can see that all other 16X slots (black excluded because it is only 4X and runs to the PCH) run through the NF200 bridge. So as soon as you put another card in any of those red slots, the NF200 becomes 'activated'. What I mean by 'activated' is that it steals 8 lanes that go to the CPU...from the top slot. The top slot now has access to 8 lanes, and the NF200 has access to 8 lanes...for a total of 16 lanes, because that is all the CPU has.

That should be straight forward...if you have two cards in red slots, the NF200 has 8 lanes and the top slot has 8 lanes. The rest here is just what happens up to the NF200 and should be ignored if you are comfortable with the explanation above.

It gets a bit tricky here though because GPU-Z will show cards #3 and #5 as 16X if #4 and #6 are not populated. But these slots only operate at 16X up to the NF200 bridge. At that point, they then get combined and sent to the CPU over the 8 lanes the NF200 has commandeered from the top PCI-E slot. So the CPU is still just being fed at 8X from the NF200 bridge, and 8X from the top PCI-E slot...but #3/#5 run at 16X up to the NF200 bridge. If you then have a card in #4, slot #3 gives up half of its lanes up to the NF200 bridge and both #3 and #4 run at 8X up to the NF200 bridge, which then get combined onto the single 8X lane going to the CPU.

The same is true for slot #5/#6 if they are both populated.

Just remember, the LGA1156 socket only has 16 PCI-E lanes going to the CPU so that is all that it can possibly accept at any time. regardless of how many NF200 bridges are on the board. HTH

If you want to see or understand how the NF200 multiplies lanes, look up 'Mulitplexing' or 'Multiplexer' on the web.

later
Raja

mtbiker033

this is really confusing, P55 has only 16 to the CPU, so no matter how you slice it, it boils down to x8/x8 in SLI right?  Btw, I can tell you from experience with GTX260 SLI that x8/x8 doesn't hurt a bit and is faster than x16/x16 on my previous 775 board.

and those 16 lanes go straight to the nf200 on nf200 equipped boards and multiplies it by 2.

So what is the benefit of the NF200?  Is it purly the ability to physically install more GPU's where the NF200 is simply manageing the data going to the CPU? 

A nf200 chip on a evga p55 classy is connected right to the cpu x16 pci-e controller the nf200 chip than takes those 16 lanes and multiplies them to 32 giving you 32 lanes of bandwidth 16 of those lanes from the nf200 chip go right  to the first pci-e slot then 8 of those lanes bridge to pci-e slot 2 giving you true 16 to slot one if a second card is not installed or x8 if two cards are present the remaining 16 lanes from the nf200 chip go to your third pci-e slot hope this helps.

I want to run three GPU's not in an SLI environment.  
 
How would the lanes be distributed with a NF2000 board?
 
Thanks

e-money

if its for your p55ftw board in your sig it would be x8-x8-x4

x8-x4-x8 actually.