This blog entry I write because I feel it's necessary to raise a point about overclocking processors and GPUs. This is hot off another late night attempt to coax just a little more love from my Core 2 Quad Q6600, which ended in tears, misery, and reinforcement of the status quo. I'm not overclocking on an ECS board either, here, so we're clear — I'm using an X38-based Gigabyte motherboard. Simply put, my G0-stepping Q6600 isn't happy going over 3GHz. I shoot for round numbers and at one point was able to get it up to 3.2GHz, but now we hang out in the magical land of 3GHz, using a multiplier of nine and a 1,333MHz front-side bus. This seems to be the best compromise, since I don't have to ramp up my CPU fan just to keep temperatures down and thus defeat the purpose of having a silent machine.
My Visiontek Radeon HD 4870 512MB fares no better. It's rock stable at stock speeds, but a journey into overclocking land is always fraught with disappointment. The GPU can't handle any more than the stock core speed, and overclocking the memory on the RV770 chip is, in my experience at least, somewhat pointless without a good core clock to match. When I tested my 4870 at a core clock of 800MHz, it felt measurably smoother than at stock speed, which was great...if I only intended to game for maybe five or ten minutes. And this is under Unreal Tournament 3; imagine if I'd thrown Crysis at it.
I bring these things to your attention because if you Google overclocks on either of these chips, you'll see far better results than mine. People on forums will talk about getting their G0-stepping Q6600's all the way up to 3.6GHz and better, at 1.25 vcore or lower, and it's Prime95 stable, we swear! The fact is that overclocking is a massive crapshoot. It's the kind of thing that we warn you about in our CPU reviews, in every overclocking section, and other sites will warn you about, as well. Overclocking is incredibly unpredictable. When the Phenom II was being promoted, AMD was shouting about how the chip could probably hit 4GHz, but our samples only hit the neighborhood of 3.5GHz—still a good overclock, but not the kind of angels-singing-from-the-rafters overclock for which one might hope.
If it seems like I'm beating a dead horse, it's only in an attempt to pound into steel what some consumers still don't fully understand. You read about these crazy overclocks, or even a legit review will get a pretty good one, so you'll go and buy the processor. Lo and behold, you don't get anywhere near what you heard about. That's the nature of the beast, so you'd better buy a processor that's going to be good enough for you at or close to its stock clock speed.
Likewise, overclocking isn't an exact science. We have an excellent guide on overclocking here, written by Geoff, that can at least get you started, but there's a lot of tweaking that you may have to do, and you also have to decide just how you define "stable." For me, for example, "stable" means rock stable. I do heavy high-definition video editing on my desktop, which has been very heavily designed and tuned specifically for that task. An overclock with "good enough" stability isn't going to be good enough for me when Adobe Media Encoder slams all four cores and redlines them for hours on end. Your standards may be a bit looser than mine, but an overclock could easily involve tweaking timings and all kinds of esoteric settings in the BIOS to reach the kind of crazy heights some of the chips on the web hit.
Overclocking video cards becomes even less exact. The "old standby" for video card stability testing these days seems to be Furmark, with ATITool running a close second, but I've had ATITool green-light overclocks that have been tested for over an hour and promptly cause crashes when the card enters an actual game. The overclock testers in software like ATI's Catalyst Control Center and Nvidia's System Tools, formerly known as nTune, have seemed even less reliable. Catalyst Control Center is all too happy to let me overclock to 800MHz on the core, "stable," and watch the card choke in actual gameplay. Nvidia's System Tools had the same issue in my laptop (Asus X83Vm-X2), where my immensely overclockable GeForce 9600M GS nonetheless had higher-clocked confirmed stability tests than bore themselves out in Left 4 Dead. For what it's worth, though, the 9600M GS does purr along quite happily at near-9700M GT speeds.
I'm not sure exactly how this is going to go over, how successful the warning will be. We're certainly past the era where the Barton-core AMD Athlon XP 2500+'s quick tweak to 3200+ speeds was considered a major overclock, especially if I'm disappointed in only scoring a 25% overclock off of my Q6600. The vast majority of chips on the market these days do have some pretty impressive overclocking headroom in their designs, but these overclocks aren't guaranteed. They can be facilitated, as AMD does with its Fusion Overdrive utility, but they can't be guaranteed by any stretch of the imagination. You may wind up being the unlucky sucker with the Phenom II X4 that only goes up to 3.2GHz or the Core i7 that only hits 3.4GHz at most. I do say "unlucky" with tongue planted firmly in cheek, given how crazy some overclocks can be these days, but we've also gotten to the point where the increased headroom has brought increased expectations with it.
All the forum posts in the world aren't going to change the fact that you're gambling when you bet on overclocking your next processor. The ATI Radeon HD 4890 may have been designed to hit 1GHz, and some partners are even releasing cards running at 1GHz from the factory, but that doesn't mean you can buy a cheaper one and just magically expect to hit 1GHz.
Buyer beware: you may get exactly what you paid for.