Running two different spec devices on one IDE channel is bad...

Has anyone got a link to a site that verifies this theory? I've seen this posted countless times, where people believe that running two different spec devices on one IDE channel will drop the transfer rate for both devices, to that of the slower one. Unless the transfer is between the two devices, i honestly can't see any reason why the IDE controller would not be able to address each device at its correct rated speed. If you were addressing both devices at the same time, then i could see that the available bandwidth may cause some slowdown, but if you had an ATA100 harddrive and an ATA33 harddrive on the primary IDE controller, and you were only accessing the ATA100 drive, it should transfer at its maximum rated speed? <img src="confused.gif" border="0">

<a href="http://www.anandtech.com/showdoc.html?i=1237&p=1" target="_blank">http://www.anandtech.com/showdoc.html?i=1237&p=1</a>

This is a useful link, but there are many pages of tests to wade through, and it's initially confusing as the subject doesn't seem to be being addressed. It's only at the very end that having a non-ATA66 device on the interface is assessed.

This is basically the conclusion reached:

"You can also see that you will not realize too big of a hit, if any, on performance by placing a Ultra ATA/66 hard drive and non Ultra ATA/66 device on the same IDE channel."

Quite simply it depends on the controller and how disk transfer requests are wrapped using dma. So you can't say for sure that differing speeds on the same ide channel will have a speed impact, but mostly they do.

If you only have one channel for data to move through and two different speed devices on the same channel how do you arbitrate? Different southbridges do it differently by compensating for speed, timeslicing or simply fifo or whatever but however you do it chances are the slower device will take longer.

Now you could go and look up how every southbridge known to geekdom works, or you could just go with the rule of thumb which says that when possible don't mix device speeds on the same ide channel.