yes, it can cause problems.
Yes, overclocking can create issues with dual card setups (both CrossFireX and SLI.) You have to be careful about PCIe timings.
Well, I’ll be damned! Lloyd and Rei are right!
Setting BCLK to 100 (default) from the “auto extreme overclocker” setting of 103 and not overclocking does seem to have worked. I was able to play several BF2:BC rounds with no issues, though I did see a hint of desync at one point, it went away by itself.
Just Cause works fine as well, that was just a corrupt install that Steam autorepair fixed.
I still can’t get BF3 beta to work with this, though – game loads fine, but the minute I ALT+ENTER to make it fullscreen, kick to desktop.
So: yay! I’ll continue messing around with this for a bit.
Interestingly, syncing the memory and core clocks of the two cards (they were subtly different at 850 Mhz / 875 Mhz and 1200 Mhz / 1250 Mhz) in Catalyst Control Center seems to have allowed me to do ALT+ENTER in BF3 beta successfuly.
A finicky beast, this CrossFire.
I’m never going to bother with SLI or CF. Threads like this and a couple friends who tried/have it are why. Also the micro stutter issue documented on techreport is another good reason as well. If I’m having trouble running a game I’d rather just turn down settings.
Pretty much. You have to remember it’s basically a marketing feature. The sort of people looking for that sort of performance are near-universally technologically sophisticated and willing to spend where it’s possible, even if a little tricky.
So it gets some resources, so it can exist, but not enough to make it smooth waters.
Well, after
- the latest n’ greatest clean driver install
- fixing the PCI clock issue I had with overclocking
- making sure clock and memory speeds are synced on both cards in the ATI overdrive settings
… this 5870 crossfire system is working well, and I can recommend it.
I did have to go back and optimize a few things about airflow in my Corsair 600T case, such as snipping the front 200mm fan grille (rear was already snipped), taking out the second topmost set of removable drive bays to provide more room for the front fan, and hooking up the built-in fan controller (with handy top adjustment knob) for the front and rear fans for a little more “oomph” in getting rid of the heat from two 5870s. It’s still very quiet.
http://www.codinghorror.com/blog/2011/07/building-a-pc-part-vii-rebooting.html
Now it’s humming along quite happily and takes any game I throw at it… finally.
So Crossfire clearly DOES work but I’d think hard about overclocking (and airflow) with a Crossfire system just to reduce variables if nothing else.
The overclocking thing is primarily a Sandy Bridge issue-- the Sandy Bridge base clock is used for just about everything in the system, including PCIe*. If you have an unlocked Sandy Bridge chip, you shouldn’t have any trouble with only changing the multiplier (I haven’t).
I agree. Forget about bushing BCLK. Just get a “K” processor – they’re certainly cheap enough now – and up the core clock, if you really want to overclock.
I told you guys, the ASUS e-z overclocker decided to change bclk from 100 to 103, not me.
We’re not providing advice so much as information for those who’ll walk down the same path in the future. :P
old, single ATI 5870 power results (4.4 Ghz CPU)
Idle at Windows desktop: 128w (multiple monitors)
Prime95 full load: 255w
video stress (game): 210w
Prime95 + video: 332w
new, dual ATI 5870 results (4.0 Ghz CPU)
Idle at Windows desktop: 120w (single monitor) 136w (multiple monitors)
Prime95 full load: 215w
video stress (game): ~400w
video stress (furmark): ~530w (!!)
Prime95 + furmark: ~620w (!!)
Not exactly apples to apples since I lowered the overclock of the i7-2600k CPU a bit to compensate. We have obscene amounts of CPU power anyway, that’s not even remotely the bottleneck in any current or new game.
But, almost exactly a doubling of power consumption when gaming, which I guess isn’t unexpected. That furmark thing is a monster, though. Really scary. No game loads the system as much as that damn synthetic benchmark does.
edit: the idle power results are quite good. Apparently per tech-report:
Additionally, the second 5870 board in CrossFire multi-GPU config can go even lower (than the first board), dropping into an ultra-low power state just below 20W.
I don’t know if it’s the same across all cards, but my second 6970 will, when not needed, shut down to the point where it no longer reports any clock speeds, fan speeds, or temperatures to the Catalyst control panel.
Right – it dropped from 148w to 137w after a while (I edited my post, above, to reflect the new multiple monitor idle power). The drivers seem to be very smart about turning the secondary card to the lowest possible power levels when not in use.
I’d say this whole thread detailing the hassle of SLI/CF pretty much explains the case against it.
I’m not sure that’s a valid conclusion to draw here, if only because I can say that the only problem I’ve had with Crossfire is a defective fan on one card.
wumpus’ situation was kind of a perfect storm: his cards came running at different clock speeds (which can cause artifacts), and his overclocking tool and processor architecture did not make it clear that the PCI clock was also changing.
FWIW crossfire has been very reliable and stable for me once I ironed out the above issues. I’ve been throwing tons of games at this config since then and everything just works.
But, as mentioned
-
make SURE you get cards that have the exact same clocks
-
turn off any CPU overclocking just to be safe when going from single to dual GPU
Shot of the interior with the crossfire setup. Note that:
- top 200 mm fan was removed
- front and rear fan grilles tin snipped (this is essential)
- side panel covering this area also has foam on interior to absorb fan noise
Mostly all I can hear is the white noise of whooshing air, and it’s nominal at windows desktop. A bit more noise when gaming and the two ATI card fans spin up of course.
What are your temperatures under load like, as a matter of curiosity?
Nice setup. Very neat, I like that.