Crossfire HD4870-X3 [4870x2/2GB + 4870/512MB]; Anyone here run it?

[cusideabelincoln]

Originally posted by: jaredpace
i think p35 vs. x48 with crossfire is game dependent

Look at the differences per game:

crysis

devil may cry 4

sup com

wic

ut3

Supcom & DMC4 show differences with every motherboard. Have you tried the 4870 CF rig for x48 vs. p35 on different games?
edit: btw, it looks like you get more performance with 8x + 8x vs. 16x + 4x

I think your linking to the pictures overloaded their server, because I cannot view them.

 

[hclarkjr]

thank you for the info apoppin :thumbsup:

 

[Tempered81]

sorry if those links aren't working, just look at the graph at the bottom of the page:

http://www.legionhardware.com/document.php?id=761&p=3

It says, "% Boost over a single HD4850".

In DMC4, Crossfiring two 4850's resulted in this:

Intel P35's 16x Pci-e 1.0 + 4x Pci-e 1.0 yields a 53% increase by using crossfire
Intel X48's 16x Pci-e 2.0 + 16x Pci-e 2.0 yields a 95% increase.

That's a significant difference.

 

[apoppin]

You need to realize that is for one game

CF-X3 drivers are immature

However, there is definite performance increase moving from a linked secondary 4870 in the 2nd 4xPCIe CF slot to a 4870 in a full x16 slot
- sometimes it is a lot, sometimes not so much; and it does not seem to matter much if it is a 1GB or 512MB 4870 paired with a 4870x2 as the primary adapter

it also does not seem to matter if it is PCIe 1.0 vs 2.0 for either slot

 

[cusideabelincoln]

Hmm, since you've done so many tests, it would certainly be interesting to see at what PCI-E bandwidth a single video card like an HD4870 is bottlenecked...

PCI-E 4x 1.0
PCI-E 8x 1.0 or PCI-E 4x 2.0

Looks like 16x is enough no matter the bandwitdth, but PCI-E 8x 2.0 would be an interesting comparison.

 

[apoppin]

i think you will be able to infer that from my testing with CFx3. The performance delta with the secondary linked adapter 512MB HD4870 in the 16x CF slot over one in a 4X slot.

Putting a HD4870 in a 4x PCI slot would be akin to sticking it in an AGP slot. It does starve the bandwidth a bit. i know from my tests with 2900p [oc'd to XT] in the 4x slot scaled nearly as well as having 2900XT CF in the 8X+8X MBs of the time; so this generation strains the 4x slot too much for it to be the primary slot .. and it even lacks for 16x+4x CF compared to the full 16x+16x

and i don't have a 8x+8x CF MB to test. i understand it is a decent compromise - probably a bit limiting for a X2 in the 8x slot, however - my guess