Conventional wisdom is a funny old thing. If one person says something and three people repeat it, then it's true. This maxim applies in full force to the world of PC gaming. One of these common sense laws which has played out like a broken record over the past couple of years is the answer to the question: which operating system is best for gaming?

The answer of course is Windows XP, right? I mean, that's what people in forums all over the planet will tell you, so it must be true. I pondered this question myself a few weeks ago, and it occurred to me that I didn't actually know. Being a PC hardware reviewer, I hate not knowing stuff, so rather than taking everyone's word for it I set off to find the answer myself.

Testing methods

A recent poll on our hardware forums showed that over 80% of GP citizens use one of the following three Microsoft operating systems for gaming: 64-bit Windows 7, 64-bit Vista, and 32-bit XP. So these are exactly the OS's that I will use for comparison. Fresh versions of each OS were installed along with all available updates being applied.

To test the performance of each OS, I have selected a suite of five canned gaming benchmarks to provide repeatable and consistent results for comparison. All game tests are run at their highest in-game settings (except for FarCry 2 which is at its second-highest), first without AA (anti-aliasing) enabled and then with moderate AA enabled, and the results taken from the average of three runs rounded to the nearest whole number of frames per second (FPS). DX9 results where available for all tests obviously, and where DX10 was available I also used that - Windows XP does not support DX10 however so that OS will be excluded from those particular results.

I also wanted to get a feel for multi-GPU performance so I ran all tests again with a second GPU added, to see if Crossfire yielded better results on any particular OS.

Test setup

• OS1 - Windows 7 Home Premium, 64bit, Build 7600
• OS2 - Windows Vista Home Premium, 64bit, SP2
• OS3 - Windows XP Home, 32bit, SP3
• CPU - Intel Core i7 965 Extreme @ 4.00GHz, HT off (supplied by PlayTech)
• MOBO - Asus P6T Deluxe V2 (supplied by PlayTech)
• RAM - 2x2GB G.Skill Trident @ DDR3-1600 CL7 (supplied by PlayTech)
• GPU - XFX Radeon 4870 512MB x2 (supplied by PlayTech)
• Drivers - Catalyst 9.7
• Screen - Samsung 206BW 1680x1050 LCD
• HDD - Western Digital Blue 640GB SATAII

One proviso worth mentioning at this stage - due to technical limitations, 32-bit operating systems such as the version of XP used here today can only address between 3 and 3.5GB of system memory. Games don't normally use that much but just to even the playing field a wee bit I took one of the sticks from my 6GB kit out of my system so that the 64-bit OS's only had 4GB to play with.

DX9 Performance

So let's get down to business, starting with the DirectX 9 results on a single 4870:

Why are most of the green lines shorter than the red and blue lines, I hear you ask? Well, I'd hate to shatter your firmly entrenched perceptions of reality my friends, but that is because both Windows 7 and Vista actually outperformed Windows XP (in most of the tests and settings that I've chosen to use at least). The slightly higher amount of RAM didn't help them either - during the most memory-demanding game (Crysis Warhead, of course) just under 3GB of system memory was swallowed at its peak.

To be fair though, XP does manage to keep up most of the time - Resident Evil 5 seems to have the only significant drop in performance.

Now let's see what happens when we throw the second 4870 into the mix:

Here the difference is even more pronounced - a 2 to 3 FPS drop between XP and the other two operating systems has turned into 12 to 16 FPS drop once Crossfire is enabled. Aside from an anomalous surge in performance in Left 4 Dead with 0xAA, the poor old 32-bit OS falls behind in every single game, most notably Crysis where it doesn't improve at all from the single card test.

Windows 7 also seems to be edging away from the other two OS's, providing better Crossfire performance than Vista. Regardless, all operating systems see over 50% performance increase when adding the second card, not bad considering that in the early days of Crossfire this increase was somewhere around the 20 - 30% mark.

Next page - DirectX 10 tests...