In case anybody is wondering if dual PCI-E 2.0 8x is ok for crossfire performance. I have a Asus P8P67 PRO, 2600K, 8GB memory, so max PCI-E 2.0 8x for crossfire. At first everything seems to be ok, heaven benchmark, metro last light, some other games. All seem to scale perfectly. But still now and then i've had these slowdowns.... it reminded me of the AGP days where it would happen if the gpu memory wasn't big enough and it started swapping textures over the AGP bus. The more i tested the more it seemed the dual PCI-E 2.0 8x speeds where limiting the performance. The final test which convinced me this was the case was Farcry3. With crossfire (dual PCI-E 2.0 8x) there where BIG slowdowns, especially looking up the sky. With crossfire disabled but still with (PCI-E 2.0 8x) performance already a lot better, but still a slowdown is noticable. (again look at the sky) Finally i completely removed the second GPU so the PCI-E 2.0 @16x again. Now the performance is good again. I am not sure why this is happening, looking at the sky in farcry3 there is nothing to render... but still. This is all tested without vsync and the 13.12 drivers (later also with the 14.1 beta driver but its the same). Can someone with crossfire check if looking at the sky in farcry3 causes a slowdown? If yes/no what config you have?
I dont have a problem with 8x 8x sli on my mb. Maybe because your cards do not have an xfire bridge? And communicate through pci e slots? Mb is nearly the same as yours. What psu do you have? Maybe 750 isnt enough for 2 cards
These cards don't have the bridges, i believe this is part of the problem. I've used SLI and crossfire before (always with bridges) and it never was a problem. The corsair tx750 PSU is ok i think/hope; it isn't crashing and 12V stays stable under full load (for what its worth). The cards aren't overheating or downthrothling either. I might try again with just the single card in the PCI-E 8x slot... But for now its in my mining rig again. Ah resolution i tested was 1920x1080 with different AA settings.
I am assuming 120Hz? If so, then the bandwidth required by bridgeless CrossFire is about ~750MB/s @120FPS, equivalent to about 1.5 lanes of PCI-E 2.0. At the very most, if you're bottlenecked so hugely by the PCI-E lanes (and you won't be, not at all), that should drop your performance by 1.5/8 which is ~19% max. So I doubt that bandwidth is the issue here. Do you get any performance drops if you cap your FPS and, while doing so, crank up settings such as SSAA to the extreme, to make the GPUs work hard? Have you tried setting the monitor to 60Hz if it's 120Hz?
Did a lot more testing, i believe the R9 290 is just not compatible with my mainboard... or whatever. This happens without crossfire, just a single card: In the slow places (looking at sky in FARcry3 for example) i see the framerate halving if i insert (whatever card) into the second PCI-E slot (so speeds are 8x instead of 16x). After testing i conclude that even when when the PCI-E 2.0 is @full 16x speed the card slows down at places where my GTX770 doesn't do this at all. (we are talking 40FPS vs 135FPS). There either is a big driver issue which affects a lot of games, or there is a compatibilty issue with these cards and my mainboard. Thats why i will try to do some testing with a PCI-E 3.0 mainboard soon.
Just note, crossfire bridge can give 2560x1600@60Hz at best. For resolutions/frequency over this it's better to send it via PCIe unless you want some frames to be lost. I believe amd may have changed their driver code to make crossfire bridge disabled for resolutions higher than mentioned anyway.
I use CrossFire with two bridges for 2560x1440 @ 110Hz. Up to 80Hz or so should work reliably with one CrossFire bridge even. No frames lost, confirmed.
No computed framerate loss, that's for sure. But fcat proved that on 4k TV every second frame @60Hz got dropped due to Crossfire on HD7990/2x7970 on older drivers. That is why I presume AMD disables CF bridge and use PCIe even on older generations once you plug in multiple screens or go over certain resolutions+refreshes. Not that it would matter to me if data go over CF bridge or PCIe. As CF bridge tech has to be dropped for future standardization. It's not used for APU anyway so driver support was around for long time.
It will have big time slowdowns in certain games, while others it won't. PCI-E 2.0 @ 8x is the same speed of 3.0 @ 4x. These charts are from 2 years ago with a 7970 @ stock, probably better drivers have improved some performance and if you overclock the % of bottleneck is more apparent. http://www.anandtech.com/show/5458/the-radeon-hd-7970-reprise-pcie-bandwidth-overclocking-and-msaa
It's also worth noting that FarCry 3 has always been fairly broken with amd cards, ran horribly on my 7970's in crossfire, was much smoother with just the one card.
Just tested with an MSI Z87-GD65 Gaming board and the problem isn't there. Also adding another graphics card so the PCI-E 3.0 is @8x doesn't hinder performance at all. I just reinstalled the ATI to my pc, and the problem is ofcourse there again. Will try with a new windows7 install to see if it makes a difference.
The new method AMD employ on R9 allow them to saturate pci-e bus, so bigger the bandwidth the better basically. Also people linking charts to 1 GPU tested across varying pci-e lane speeds an saying not much of problem...yes for 1 GPU it's not much of problem, but once you go 2 or 3 GPU's you'll see big differences from pci-e 2.0 to pci-e 3.0 as much as 40-50%
Your original post was about PCI-E 2.0 @8x, which will hinder it, glad to hear PCI-E 3.0 @8x won't hinder it.
But it shouldn't not like it does at the moment. 140FPS vs 40FPS is a big difference! Just tried a clean install of windows7 x64, but that didn't fix it.
Well how much bandwith does the bus use to transfer the xfire signals? Could be a lot, also a stock 7970 is heavily limited in certain games on 8x pci-e 2.0. So yeah thats a bad situation for those cards to be in.
Its not even a crossfire problem. It happens with just a single card active! Overal it seems to run fine, but there are parts in games where the problem is noticable. Farcry3 sky @night for example, certain parts in BF4, NFS rivals, etc. For example looking at the sky in farcry3 @night: Single R9 290 in PCI-E 2.0 16x: 35 to 40FPS Single R9 290 in PCI-E 2.0 8x: 20FPS Single R9 290 in PCI-E 3.0 16x: 140FPS Single R9 290 in PCI-E 3.0 8x: 140FPS Single GTX770 in PCI-E 2.0 16x: 135FPS Single GTX770 in PCI-E 2.0 8x: 135FPS Then there are the benchmarks/games that run just fine, also with crossfire. Heaven benchmark and metro last light show perfect scaling, even with the PCI-E 2.0 8x slots. PS: this is all with the DX9 exe of farcry3
There are no dropped frames. Computed or not computed. No dropped frames. I can VSync at 110Hz and there is no dropped frame. So can all the Qnix QX2710 110-120Hz users on that particular thread. AMD have supported up to 2560x1600 over CF bridge but they did not specify the refresh rate limit. Given that 110Hz @ 1440p is unofficial, and works better with two bridges, it would seem that two bridges are being used, as many can confirm. You're talking theory, I'm talking practice. If there was a single dropped frame I would be raging about it. But there isn't.
overclocking the PCI-E bus from 100 to 105 mhz (keeping cpu @ around same speed), increases the mininum FPS in farcry3 by 3~5% !!!