I've used the beta and near final versions at my former workplace with the following configurations:
K6-something 400 MHz, Matrox G200, ATI Rage Pro, Win2000 and Win98
PII 266, Matrox G200, ATI Rage Pro, Win2000 and Win98
Pentium 200, Riva128, Voodoo 2, Win98
Pentium III 450 Riva128, TNT2, Win98
We also had a hardware compatibality lab, that tested it a wide range of hardware, using D3DX.
I am currently using the final version with:
Pentium II 350, MPact3D, Win98
As for it switching to software, have you tried enumerating the devices and see that it's recognizing the hardware?