Just a quick question. I'm hearing that the X58 can run at 16/16 sli mode and that the P55 can only do one at 16 or 8/8.
Everyone says there is a bottle neck with the 8/8. What I'm wanting to know is, how do you calculate the bottle neck? Surely there must be a way of working out "well these two cards can go in sli and won't reach the 8/8 bottleneck".
Does anyone have an answer for this one?
I've been researching a bit more and have had further thoughts. The PCI-E lanes are used by other devices on the board, not just what you plug into the slots right?
For example, if I had something running on SATA 3 and USB 3.0 reading from an SSD, that would chew up a couple of lanes would it not?
Do you know of a way to calculate the throughput a card will need for the PCI-E links?
To expand on this some. PCIe unlike PCI is a point to point (non shared) bus. PCI worked like you described. Plugging in two cards meant they had to share the total 133MBps that PCI can deliver. PCIe is different in that each slot gets its own bandwidth and doesn't need to share it at all. That 16x slot has 16lanes no matter what. (assuming 16 lanes were hooked up to it.)
What does get shared so to speak is the limit of lanes that a chipset can do. Lets pretend that chipset X can handle 30 lanes of PCIe. You obviously can't do full speed CF/SLI as you'd need 32 lanes to do that. What you can do is 16 lanes for GPU(s), 4 lanes for any chipset communication, 1 lane for any gigabit LAN, 1 for each of your 3 PCIe 1x slots, and perhaps a 4x slot like the P35 chipset had. This takes care of the 30 lanes you had.
Alright so that gets the sharing resources question out of the way. Lets look at this scenario;
Say for example you have a really great, top of the line graphics card. You slot it into one slot and it runs fine in 16X (because you've got all those lanes for it to use). Now you put 2 of those cards in a P55 motherboard. So now its 8X for each card. If that graphics card was pumping out more than 5gts (4gbs with the overhead removed). Then you've got a bottle neck, because 8X can't pass the data through fast enough.
So we all know what the PCI-E 2.0 lanes are capable of (500Mbs). So 8 lanes X 500Mbs = 4Gbs, 16 lanes X 500Mbs = 8Gbs
So now to determine the bottle neck, we need to know how much a card pumps out on those lanes, its maximum throughput.
I'm having ENORMOUS trouble trying to find an answer to how to calculate a cards throughput. At the moment I'm trying to get in contact with a Nvidia or radeon tech to see if they can help me.
Does anyone have any ideas on how to calculate the throughput a graphics card pumps out?
Then you've got a bottle neck, because 8X can't pass the data through fast enough.
Says who? Just because you have 8x instead of 16x doesn't automatically mean your going to have a bottleneck. I doubt the techs will be able to help you as there is no formula you can run to see if you'll have a problem. As a matter of fact, its not the card that "pumps out on those lanes", but the CPU thats feeding them, along with requests from memory. Toms and other sites have run tests on 8x/8x vs 16x/16x. Look at the review of the i5 series. There is some minor speed loss with the fastest cards, but its not much to worry about.