good read of truth of 16x 4x crossfire pcie

Good stuff, thanks for that.

Still, I would've liked to have seen, just for fun, how much the numbers would've changed if they used 2 multi-GPU cards instead of 2 singles. That's where the extra bandwidth would've actually meant something.
 
It would be interesting to see if overclocking the cpu would make a difference.
 
yeah i know if they used 2 6990 would of been good to see, but i was really suprised to see the numbers on singles cards.
 
This was the main selling point of x58 boards having full 16x by 16x
 
I have 16/4. I'll be crossfiring.

i dont understand how you will be running crossfire at 16x by 4x. the 1155 cpus have 16x pcie built into the cpu, yet your motherboard mfg says it runs 16x by 4x. how is that even possible? if anything it would be 12x by 4x. is the second gpu running off of the north bridge?
 
Last edited:
I think most of the stories about 4x killing the performance of xfire was when the pcie 1.x standard was out. I'm pretty sure that pcie 2.0@4x = pcie 1.x@8x, which is pretty acceptable, at least thats how i understood it.

When i had xfire 4890 there was no benchmark diff between the 2nd card running in 4x or 8x. (and yes it had higher #'s than a single card)
 
i dont understand how you will be running crossfire at 16x by 4x. the 1155 cpus have 16x pcie built into the cpu, yet your motherboard mfg says it runs 16x by 4x. how is that even possible? if anything it would be 12x by 4x. is the second gpu running off of the north bridge?

The additional lanes are coming from the PCH (USB 2.0, PCIe x1, SATA, LAN). The main issue I've been hearing is that if you Crossfire x16/x4 off of a P55 or P67 board is that you run into micro stuttering. Haven't personally tested it myself but it would be something I'd like to hear from the reviewers. Another thing is if the minimum fps is greatly affected by the bandwidth differential.
 
At the resolutions that were used in the test, x16/x4 vs. x8/x8 will only make a small difference.
At Eyefinity resolutions the difference will be much more noticeable.
 
I had always assumed x8/x8 was better. It's good to know I was wrong since I have a x16/x4 mobo
 
how about they use cards that scale well in xfire? like 6xxx or 57xx instead of 58xx which are known to be gimp in xfire :p
 
Well,. the [H] did these a year ago, same result I think... can see here for 8X8 and here for 4X4. New format is not letting me look at the 4X4 do, hope they fix it up soon.
 
Well,. the [H] did these a year ago, same result I think... can see here for 8X8 and here for 4X4. New format is not letting me look at the 4X4 do, hope they fix it up soon.

the 8x8 is good too, since it does multi-monitor resolution. There isn't much of a difference still about 5-7% depending on game, however 4x4 might take a bigger hit with multi-monitor res.
 
The additional lanes are coming from the PCH (USB 2.0, PCIe x1, SATA, LAN). The main issue I've been hearing is that if you Crossfire x16/x4 off of a P55 or P67 board is that you run into micro stuttering. Haven't personally tested it myself but it would be something I'd like to hear from the reviewers. Another thing is if the minimum fps is greatly affected by the bandwidth differential.

P55 has a bigger disadvantage than P67 in using the PCH to run a graphics card (16x4 configuration) because of the following:

P55 PCH has PCIe lanes that run at 1.0 speeds! That means the 4x slot is really like a PCIe 2.0 2x slot, and can constrain performance.

P67 fixes all that because:

P67 PCH has PCIe 2.0 speeds on all lanes, and has 4 available to dedicated to a video card (the other 4 are needed for IO).
The DMI bus bandwidth is increased to feed those new 2.0 lanes. There is enough DMI bus bandwidth to feed a 4x PCIe 2.0 video card.

So, in conclusion:
16/4 (or 8/8/4) on P67/Z68 = GOOD.
16/4 on P55 (without NF200 to help) = BAD.
 
Saw a Linus gtx 580 sli with 16 x 2 vs 8 x 2 and there was only a. 01 difference.
 
Back
Top