For RAID controllers that come integrated on motherboards it's usually fingerquotes hardware. Meaning that the OS driver does all the heavy lifting. The distinction doesn't really matter for RAID0/1 though because there is no parity data to generate, and conveniently that's all they usually support. It would only really matter for RAID5, and in that case software RAID5 almost always trounces hardware RAID5 because the CPU is much faster than whatever embedded processor is on the card.

And yes SATA-2 is pointless at this point because no drive even comes close to sustaining 150MB/s let alone 300.

Thanks! So that probably means in Linux, I should just do an all-software solution, rather than messing with whatever integrated RAID controller is on the board?

software RAID5 almost always trounces hardware RAID5

That is the exact inverse of my own experience. Please explain how a generic CPU, busy with OS tasks and on the far side of a PCI Bus, beats custom ASICs directly on the raid I/O controller.

Knave: IMO & IME, hardware raid soundly beats software raid in every dimension except price. Disrearding performance questions, by doing it in hardware you are abstracting the redundancy operations from the host OS - which means that you do not need a bootable system to be able to recover from failure.

And single-spindle speed isn't the only consideration wrt 150 vs 300. Maybe one spindle can't saturate that, but multiple spindles in a stripe set can. Granted, you're not building that, but since you asked...