Just looking at it makes me want to punch something.
Backstory
If you didn't know, I just built a new computer about 3 months ago. I decided to go for the works; Sandy Bridge processor, 8GB of RAM, huge CPU cooler....and dual graphics cards. Now, I haven't put a million systems together in the past; in fact I've done relatively few upgrades to the PCs that I've had as well. Because of this, I've got no brand loyalty at all to either ATI (RIP) or Nvidia. When I was looking to put this system together, the parts I chose were simply the best from a price/performance standpoint. I wanted to spend about $2,000 on a PC that would tear anything apart; my goal was 60+ FPS @ 1920x1200 on the most demanding games like Metro 2033 and the original Crysis.
I based that criteria on a trend that I've been noticing in the PC world: regression. No one is trying to push the boundaries like those games did. No one is building a game that is impossible to max out and would remain that way for three years. Most companies are producing titles that are multiplatform, and most of the PC versions are designed to have an image quality comparable to their console counterparts. Take Crysis 2, for instance: it's a beautiful game but one that doesn't punish hardware or look as draw dropping as its predecessor. I can chalk up the improved performance to Crysis' poorly optimized code, but there's no denying that, at best, Crysis 2 looks as good as its 3 year old father, and never any better. With that in mind, I set these games as my benchmarks. If I could get hardware that handled those two at my required level of performance, I knew I'd be in the clear for at least two, perhaps even three years with little to no reduction in image quality.
Enter the 6950.
Double Your Pleasure, Double Your Fun?
Codename: "Long and Strong"
I wasn't getting the numbers that I wanted out of a single card. It just wasn't happening. The 570 and 580 from Nvidia both wimped out when hit with either Crysis' Enthusiast mode (why in the HELL they decided that normal, high, and very/ultra high needed to be renamed will forever be a mystery) or Metro's tessellation-heavy maxed out firepower. And the (at the time) brand new lineup from AMD wasn't what people were hoping for. They positioned their highest end card, the 6970, against the 570 and not the flagship 580. This meant they posted even worse numbers, and sometimes were even passed off by last generation's GTX 480. But something interesting happened when you stuck two of them in the same machine. You see, AMD spend their time making optimizations to their Crossfire setups, raising them from the level of "totally worthless" to "kick-ass but impractical". And seeing as I wasn't building a machine for the sake of practicality, that was all right by me. I'll just let the numbers do the talking:
Fuck. Yes.
Now we're getting somewhere. $600 worth of hardware competing against $1000+. The scaling is just extraordinary, actually reaching double in some cases. 75% efficiency was unthinkable with last generations Radeons, let alone 90%+. The pricing was perfect as well; two 570s were just out of my budget, two 580s even more so. And with the ability to flash the 6950s and more or less turn them into 6970s, it seems like a fucking home run. So I bought the cards, picked up the rest of the parts, installed them all without a hitch (well, one or two hitches. You know what I mean.) and everyone lived happily ever after.
Right?
Wrong
Wrong.
Dead fucking wrong.
You see, I thought I did my homework on this. I made sure I had the PSU to handle it all. I made sure that my motherboard would split the PCI lanes in a more or less logical way. I knew, I fucking knew that Crossfire/SLI setups can be temperamental. I read up on all of the potential driver conflicts and hardware issues and weird clock speed mismatches and "microstuttering" (whatever the fuck that is) problems. Most of them I chalked up to user error, or to personal vendetta from a hardcore fan of the other company. I wasn't alarmed by the volume of topics simply because I applied the old rule: You don't post to say "Hey, everything is great! I love you AMD/Nvidia!", you post when you have a problem. And you do it loudly, loud enough to drown out the few voices that did bother to post and say they were doing just fine. On top of all that, AMD spend time making Crossfire suck less. The numbers prove it....right?
Well, they weren't lying about the numbers. Firing up Crysis, laying a bunch of explosives under a jeep, hitting the detonator and watching as my computer doesn't even fucking flinch is a great feeling. Rolling through Metro 2033's tunnels with my PC hooking into my 7.1 home theater system is an amazing experience. The problem is...I don't get to do that as often as I should. My roommate puts it best: "I get 20FPS all the time, you get 200FPS every six tries". It all started so subtly: some app crashes, perhaps the display driver would have to reset every once in a while, stuff that wasn't out of the ordinary even for a single card system. I was, and in some ways still remain, able to put up with all of it. The high points are just so damn high that I can overlook whatever little faults creep in and break the illusion.
But lately, things have been going south. App crashes proceeded to blue screens. Driver resets gave way to hard locks. I found myself unable to run CPU and GPU intensive workloads without the whole thing locking up (bye bye folding, at least for now). Sometimes a BSOD would cause the fans on the cards to kick up to 100% (which I would describe as sounding "leaf blower-esq", seriously you have to yell over the damn things), a value they NEVER reach under any load. Sometimes I couldn't even open the Catalyst Control Center (the CCC is a necessary evil if you want to use Crossfire) without it pulling up a BSOD. I began worrying that it was a hardware issue, but independently testing each card (as well as getting them both to run FurMark while Crossfire'd) have ruled that out. The worst part is the god damn inconsistency of it; I could go for DAYS without a single issue, but out of the blue it would crash five minutes after boot. Sometimes it would be a hard lock, sometimes an actual blue screen would appear, sometimes with the fans going crazy. There is no pattern to it.
A new problem prompted me to post this; 2D rendering in fullscreen applications was fucked up. Despite not changing any settings, a game like TF2 or Borderlands would have very low-res textures and blocky shadows. No setting change could fix this, either in the game or in the CCC itself. I resorted to switching drivers to the new 11.4 preview that AMD pushed out a while back; that only proceeded to fuck up text rendering in Firefox (not the desktop, just in Firefox). After wasting a good two hours uninstalling and reinstalling and BSODing and booting into safe mode and trying the old drivers (now the text bug was present there as well), I finally got the damn thing back to square one, at least in terms of the text.
I could take a card out, but that would mean a) I've wasted $300 b) I won't get the performance I want and c) I would be admitting defeat. I'd appreciate any insight or even some similar tales, good or bad. Just don't mind me if I don't respond immediately; I could lock up at any moment.