Ahh, the euphoric feeling you get from hitting the power button for the first time after assembling a rig and seeing it fire right up. Even old vets get a tinge of exhilaration (and relief) at seeing the POST screen appear, representing the culmination of careful research and planning combined with careful construction. It doesn't matter how easy the process comes, or whether it's your first or thirty-first build, the machine is your creation, built to specification by you.
And then it hits you; what do you do next? You've farted around the desktop, delighted in the crazy quick boot time (which you know will deteriorate with each new app install), demolished The Murph's Frets on Fire score, and of course posted a few pics on Maximum PC's forum. That new PC feeling has started to wear off, but wait, you've only touched the surface of the building experience. It's time to benchmark!
Enthusiasts benchmark their PCs for a variety of reasons, the least of which pertains to bragging rights. While posting an ultra low Super Pi time may seem like the swank thing to do on a computer forum, I can assure you that in the real world, it's not going to impress that Sally you've been eyeballing and getting up the nerve to talk to. Instead, benchmarking should be used as a tool in fine tuning your custom rig. There are all kinds of changes and tweaks you can make, such as playing with your RAM's latencies, altering your swap file, setting up a defrag schedule, overclocking, and the list goes on. But to gauge what kind of dividends your efforts are paying off, and whether you're actually helping performance or hurting it, you need to run benchmarks and compare the results. In the end, you may not net a noticeable performance boost from where you started, but you will revel in the knowledge that you've optimized your PC to run at its full potential, which, depending on how well you did, could stave off the upgrade bug later on down the road when your new parts begin to show signs of aging.
Now that you're convinced that benchmarking can be useful and not simply obnoxious, which ones should you run? Here are some recommendations:
Futuremark's 3DMark series continues to be a popular benchmark, geared mainly towards gauging your videcoard's performance. 3DMark06 does take into consideration your processor's abilities as well, but by and large, the focus here is on the GPU. It's a free download, though there's also a full version that will set you back $20 if you're interested in the additional test and configuration options.
Another Futuremark product, PCMark05 doesn't single in on one specific area, but looks at your PC as a whole. Changes you make to any part of your PC's subsystem should be reflected in a PCMark run, and like 3DMark above, there's both a free and paid version available.
SiSoftware's Sandra suite is just an all around handy tool. It's able to tell you a wealth of information about your system's components, such as the make, model, and version of your motherboard, the BIOS vendor, your processor's vitals including it's stepping, and much more. But in addition to being a system auditor, Sandra comes with a bevy of subsystem specific benchmarks. You can test your processor's arithmetic and multi-media capabilities, benchmark your hard drives and opticals, gauge your USB key's performance, and even measure and grade your file system, all included in the free version! But the Sandra benchmark I use more often than not is recording my memory's bandwidth. This synthetic score comes in handy when playing with dividers, frequencies, and latencies, allowing you to see which tweaks are more likely to boost performance.
Super popular amongst, well, benchmark whores (there's just no nice way to say it), it's become a favorite application for the bragging rights crowd. But aside from boosting your geek cred and leveling up your Obnoxious skill-set, Super Pi, currently in version 1.5, makes for a handy benchmarking tool and plays double duty as a stability tester. The focus here is on both the CPU and RAM, with tweaks to either one affecting your Super Pi calculation times (lower is better). To gauge performance, I recommend running the relatively quick 1M test, and the much longer 32M benchmark for stability. Just don't post your scores on the forum afterwards, followed by OMGBBQ LOL mY score r0xoRs y0!
HD Tach, as its name implies, measures your hard drives performance, and can handle RAID arrays too. Just be careful with this one, as the synthetic numbers used can insinuate a much larger performance boost than you'll actually notice in a real world setting. For example, a RAID 0 array will destroy a single hard drive in HD Tach, and while RAID 0 is indeed faster, you're not going to notice a night and day difference as would seemingly be portrayed in an HD Tach run.
For all intents and purposes, single core computing is dead. Dual-core CPUs can now be picked up for under $100, and even quad-core has moved into mainstream pricing (Q6600 anyone?). And as far as our benchmarking programs go, Cinebench does a good job of measuring your multiple core processor's performance. You can test your videocard's OpenGL performance too, but I use this one solely for the CPU.
Most games come with built-in benchmarking tools, and some even have real-time FPS counters, but FRAPS works with just about any game. It's not a very good way to compare performance because, without a pre-scripted demo, you can't eliminate uncontrollable factors, such as the amount of animations on a screen, whether the path you took this time was more graphically demanding than the last, and so on. But it is a neat utility for those curious about their GPU's frame rates at any given time, as well as keeping an eye on the real-world performance.
There are other benchmarks out there, and how many or how few you choose to run is completely up to you. There's no right or wrong way to go about it, so long as you're confident that your gamut of benchmarking is providing you with sufficient information about your system tweaks. Or you may choose not to benchmark at all, but like pieces of flair, I'd recommend at least trying some out.