Gabe then went on to present some results of benchmark testing they had conducted internally. The results came from a system with the following configuration:

The scores are from an average of the 3 demos they displayed at E3, one of which is the very shader intensive “tech demo”, which may not be indicative of true gaming performance. The other two are standard game levels.

The first results are from the default DirectX9 HLSL path with full floating point precision:

Note: "RADEON 9600" is actually a Radeon 9600 PRO.

As we can see the performances are rather shocking. However, we have witnessed similar trends to these, first with the early 3DMark03 performances, and then more recently with our Tomb Raider results with PS2.0 rendering.

Valve rejigged through their results to give the rather emotive FPS per $ graph, with the latest figures from Pricewatch.

As you can see, the 9600 PRO actually appears to be the best buy for this title in terms of raw “bang for the buck” – the 9800 PRO is limited by CPU performance at the benchmark resolution of 1024x768 (as we will see later from our own further testing). According to these findings, Valve are suggesting that the GeForce FX solutions give a poor performance return on their investment.