That's the sum of HardOCP's comparison of Vista framerates vs XP framerates. According to their results, on average, XP outperforms Vista by about 25% on an NVIDIA GeForce 7600 GS and about 18% on a GeForce 8800 GTS.

It's unfortunate that HardOCP chose to use NVIDIA graphics cards, which have already been identified to have poor Vista drivers, without an AMD counterpart. It's also a pity that they didn't use their apples to apples benchmarking method to show exactly what the numbers mean in terms of end user experience. For a featured article, it's remarkably unconvincing and rushed, but better than nothing.

If nothing else, it does demonstrate that the gap narrows as GPU performance increases, supporting the general feeling that at this stage, Vista is really only worth it for the high end.

After due criticism, [H]ard|OCP have posted an extension to their article to including AMD graphics hardware in the comparison.

Not surprisingly, the Radeon X1950 XTX showed a smaller performance gap of around 10%, thanks to more mature Vista drivers.

However, their article leaves a lot to be desired. [H]ard|OCP have set out to, or at least pretended to, try and separate fact from fiction when it comes to gaming on XP vs Vista. Vista, being the most significant Operating System launch since 2000, is no small event, and the yay or nay debate is no small debate, yet [H]ard|OCP have shown what I consider to be an uncharacteristic level of ineptitude in this matter. To fudge an article is easily forgiveable; not every hit can be a home run. To revisit an article and make consistent mistakes, especially when this is such a hot topic, is rather disappointing.

• Two pieces of NVIDIA hardware were used in the first test. This allowed us to at least make a rudimentary analysis of how Vista scaled down. In the amendment, one one piece of AMD hardware was used.
• NVIDIA had two generations of hardware included in the test but AMD didn't get their latest Radeon HD 2900 XT included, you know, the card they designed for Vista.
• Graphs are great, but only when you can actually figure out what the values are. Low sample y-axis with no minor tick lines make it infeasible to actually figure out what the frame rates are. For some reason, they decided not to include a supporting table.

[H]ard|OCP are a trusted authority and this article is likely to be referenced highly. I feel rather like, if they aren't prepared to do it properly, they shouldn't do it at all; better to have no article than a misleading one.