Preferred image format for HP single-sided 3.5" floppies?

David Riley wrote: How did you reach that conclusion? The 8086 and 8080 have 20 address lines, numbered from A0 (LSB) to A19 (MSB). The 80286 added more address lines, and in "real mode", its address arithmetic was such that the resulting address could be greater than 0xffff, and thus have A20=1. Are you arguing that the 80286 should have internally masked addresses to 20 bits (presumably only in "real mode")? If so, I'd have to disagree. I think Intel made the right choice. I disagree here also. Providing a means to mask A20 seems like a perfectly reasonable thing to do. Putting it on the keyboard controller is admittedly a little strange, but then, it isn't expected that you're going to be turning the A20 mask on and off at high frequency. The big problem with the AT design had nothing to do with A20, but rather with having the only way for software to force a reset (e.g., to get from protected mode to real mode) be via the keyboard controller, which was very slow. If Intel had provided a *documented* way of getting from protected mode to real mode without a hardware reset, that wouldn't have been an issue. The processor architects at Intel assumed that once you entered protected mode, you wouldn't want to go back. Arguably that was a little short-sighted of them, but then, back when the 286 was being designed (starting around 1980) the IBM PC wasn't even on the market, so it clearly wasn't obvious that the IBM PC and its descendants were going to become the most popular general-purpose computing platform on the planet.