It would make more sense, indeed, but only if the 3.3V pins of the PCI connectors were bussed on the Mediator. There were such "new adaptors" years ago when PCI was young on intel PCs, I even could get one from an old PC. It's a PCI card shaped PCB with a low drop regulator (5V -> 3.3V) with cooling device.
BTW, a modern PCI gfx cards takes around 4Amps. Linear regulation means: heating of (5V - 3.3V) * 4A = 6.8W. This is quite a lot, if you want an efficient way you will need a switching power supply and use the 12V rail, which usually is not so loaded in Amiga systems (I have such a prototype done for Prometheus, with 98% efficiency and 6A load).
Besides, this is one of the PCI Spec 2.1 things which were not implemented in the Mediator (claimed to comply to Rev 2.1 PCI Spec).
I posted on this issue on the Mediator list, but my posting (as well as any before after Elbox took over the list) has been censored by the list owners (aka: Elbox) without comment. So I post it here, as I don't like this kind of censorship. Apparently Elbox does not like the fundamental right of free speech:
> That's right - at least it is part of the truth.
> If you look carefully into the specification
> (which is available online on several web sites,
> just google for "pci21.pdf"), take a look on
> page 142 (section 4.3.4.1 - Power requirements):
>
> "Systems implementing the 5V signaling
> environment may either ship the 3.3V supply with
> the system, or provide a means to add it
> afterward (i.e., bus and decouple all 3.3V power
> pins) to support expansion boards that require
> it, but must provide the other three power rails
> with each system."
Michael
