@voytech:
It is possible to stick an IDE controller card into the mediator, but as there are not drivers for such cards it wouldnt be usable for anything more than looking at it with a PCI-busscan program
.
Moreover it would be faster than the A1200s internal IDE, but and this is a _BUT_... PCI-cards located on the Mediator1200 will maybe be able to transfer data by themselvers (DMA) to the first 16MB (only 24-bits of addressing on the A1200 expansion connector) of the A1200 address-space... and that is presuming that the Mediator 1200 allows these kind of transfers at all. Presuming that the Mediator 1200 does allow DMA transfers to memory in the first 16MB of the A1200 address-space, the PCI-card willl in most cases only be able to access chipram. This is because most accelerators, except a few old ones only supporting 8MB fastram and pure ram expansions, have their fastram mapped in higher regions of the address-space to be able to have a continuous mapping of the fastram.
This boils down to these performance possibilities:
If the Mediator 1200 supports DMA-transfers to the A1200 address-space it will in most cases only find chipram. The DMA-transfers will be usable, but quite slow. I suppose around 6MB/Sec would be absolute maximum if very lucky, but atleast they wont strain the CPU a bit. The only problem would be if the AGA-chipset was used for graphics at the same time, then the transfers would be much slower and the graphic-operations would slow to a crawl, plus the buffers for the harddrives connected would have to be in chipram, leading to the cpu having to do filesystem work in chipram which would be slow. If a graphic-card is used this will work a bit better as the usable chipram-bandwidth increases a lot if the AGA-chipset isnt drawing anything on a screen.
If the Mediator 1200 doesnt have support to let PCI-cards do transfers to/from the first 16MB of the A1200 address-space all datashuffling to/from the card will be CPU-driven and in a sense it will put this solution on the same level as the builtin IDE of the A1200 as transfers will be very CPU-demanding. Ofcourse it will be faster, it will be as fast as the Mediator 1200 and accelerator have the possibility to communicate which should be around 8MB/Sec.
Though if there exist a SCSI-controller on your accelerator I would highly recommend using it instead as these often make use of DMA transfers, offer better performance and doesnt load the A1200 localbus.
With a Mediator 3000/4000, having a PCI IDE-card would be much better as those Mediator models definately give the PCI-cards the possibility of doing DMA transfers to/from all of the Amigas address-space which would give reasonably fast transfers with no CPU usage.
Is it possible to get programming documentation for the Mediators from Elbox?
/Patrik
