Typically, though, the controller should auto-negotiate on boot. Adaptec controllers are normally very good about this auto-negotiation...also, I believe all Adaptec controllers have a hardware termination setting (a jumper or a resistor pack), even if the setting isn't documented. I've seen that setting correct strange problems like this on AHA2940 (AIC7xxx) UW and U2W controllers.
As for the signaling, as long as you're using a 16-bit cable with both high and low bytes LVD-terminated on each end (and nothing double-terminated) you should be fine. The bus should just automatically drop to single-ended mode if there are any single-ended devices connected.
Also, to whoever started the thread (the back button is dangerous in the middle of posting
)...how long is your SCSI cable? Generally I'd worry if the thing's over 1.5 meters total (although I believe Ultra SCSI can theoretically reach up to 3 meters). Also, make sure there's at least 9 inches between each connector on the cable and no more than 6 inches between the last device on the bus and the cable's built-in terminator.
A word on TERMPWR jumpers, in case you didn't know this already: this is a bit confusing the first time you see it. "TERM POWER" settings don't control whether a device terminates the bus. Rather, if enabled, it tells a SCSI device to provide additional power on the SCSI TERMPWR line (part of the SCSI cable) so a terminator can do its job better. Enabling it may help if you have four or more SCSI devices.
Also, putting sharp creases in SCSI cable is BAD (this actually goes for any ribbon cable, but single-ended SCSI and ATA66/100 are particularly sensitive). I typically make folds in ribbon cable by rolling the fold point around a pencil or the like.
It's possible that adding the converter+device+cable is exceeding the maximum allowable "stub length" of the SCSI bus. "Stub length" refers to the length of the traces branching off from the SCSI cable to a SCSI device. I believe maximum stub length is 4 inches on Ultra/UltraWide SCSI--and the device's PCB trace lengths between connector and logic chip are included in that limit. 68-to-50-pin converters (especially cheap, cheesy ones) can add enough stub length to confuse any device that's already close to the limits.
How does the bus respond with only one device connected? Does it behave a bit better?
Kelledin
"/join #hackerz. See the Web. DoS interesting people."