Skip to main content
Home▸ Forums▸ 68kMLA▸ Dream peripherals (the return) — #64
Post #64 by trag
Source Forum68kMLA
Category68k
Post DateMon, 14 Apr 2008 - 17:59
Original URLhttps://68kmla.org/bb/threads/dream-peripherals-the-return.8591/
Post
First up, Nubus is an intelligent protocol and it requires a Nubus controller IC on each card. So you've either got to find those (hah!) or reverse engineer one, then build something to emulate it. It's hardly a "throw together", and there's been plenty of discussion around these parts as to why not.
The NuBus card requires a little logic to handle arbitration properly, but it's really not very complex from a logic design point of view. It's the kind of thing which will easily fit into a handful (or less) of cheap PLDs now days.

Second, can these ACARD chips be bought on the open market as bare chips, in small enough quantities and cheap enough to make it viable? IE, quite a lot cheaper than buying the complete SCSI-IDE converter
Don't know about that, but if they can, you probably must buy thousands at a time. I doubt that the market is much above hundreds.

Thridly are you sure it's just a single chip and not a combination of chips on the SCSIDE boards that do the magic, including say a proprietary ASIC that ACARD is unlikely to want to sell or just hand out the design? Seeing as they have this profitable line in converter cards we want to undercut.
Just two chips. The bridge chip and a Flash ROM to hold the configuration code.

All of the above being good reasons to go for a fresh PDS design.
Economics, economics, economics...

The SCSI-IDE converter requires one chip per device. So, if you want the equivalent of two ATA channels, you would have to put four SCSI-IDE chips on your card. On the other hand, there's not that much room for drives in most of the machines, so two SCSI-IDE chips would probably be sufficient.

A single FPGA large enough to do the job (two or more ATA channels, with two drives each) would cost less than $30 bought one at a time. But someone must do the skull sweat to make it work.

Performance: A SCSI interface based on the 53C80 is going to be sloooooowwww. Plus the 53C80 isn't exactly cheap any more. It's moved into the realm of "legacy" parts. If you use a 53C80 as the root of your device, you're going to be taking 40 MB/s NuBus and feeding it through 3 - 5 MB/s SCSI to reach a 20 - 60 MB/s (large dependencies on generation of IDE drive used) hard drive.

An FPGA based solution offers potentially cheaper implementation and much much faster performance. Especially if one builds it as a PDS card rather than a NuBus card, where the data and address busses are not multiplexed and the bus speed is faster.

In fairness to the originator of the 53C80=>SCSI-IDE idea, I think he was just thinking of something we could build now, without investing a lot of time in design. He wasn't optimizing at all for cost or performance.

But at that point, I think you may as well just get one or more Acard AEC-7720Us. I still have a few at $39 + ship. There is an outfit on Ebay with hundreds of them for $30, but they don't include the power cable, nor the instructions and their inflated shipping makes their price comparable to mine.

Yes, I agree it would be best if it -pretended- to be a known SCSI device so we can use existing drivers. How about the fast SCSI PDS controller from the WGS95?
This is about the only way to do it, but on machines pre-660AV and 840AV (IOW, every 680x0 machine except those two) one must basically rewrite the SCSI Manager 4.3 support and add it to the device's firmware. This is probably not trivial.