There is one peculiar thing about the results. The integrated chipsets (i865G and RADEON 9100 IGP) suffer differently from activation of the integrated graphics cores. While the i865G experiences a perceptible bandwidth reduction and higher latency, the memory controller from ATI is less vulnerable. This is probably due to the graphics core from ATI loading the memory bus less as it uses compression algorithms applied to the data pumped along the bus.

Note also that the i865PE shows higher speed than the i865G even when the external graphics card is used. Although these two chipsets don’t differ in their internal architecture, the difference is clear. I think the chipsets have nothing to do with this, for this is the mainboard’s fault. Mainboard manufacturers don’t think it necessary to optimize the BIOSes of their products on the i865G, since they are targeted at budget systems. That’s why i865G-based mainboards are slower than their mainstream counterparts: they come with average internal timings and without even mentioning the PAT support.

Now, let’s turn to applications that don’t use 3D graphics.

The Winstone tests show how fast the system is in office and content-creation applications. And they also show that the RADEON 9100 IGP is speedy enough. The main contributor to this result is the IDE controller, sitting in the South Bridge. Its efficiency matches the IDE controller from the ICH5.

Data compression and encoding performance depends mostly on the memory controller. When the integrated graphics cores are in use, the RADEON 9100 IGP performs as fast as i865G. When we install an add-on AGP card, ATI’s chipset falls behind the i865G.