Gregory's ServerIntel introduced the iAPX 432 "micromainframe" processor in 1981. Like every good mainframe, it had a separate channel processor to perform input/output (I/O). Here's my die photo of the 43203 Interface Processor chip. 🧵
 4 comments
 @kenshirriff Very intriguing thread, as in "These might be good ideas!", until I saw a memory structure diagram and I realized all that walking would be done in hardware, and how slow are they trying to be? And then I asked myself: What makes me think hardware can't walk data structures? I mean if software does it, the hardware still does it, right? What makes that slow? @kentborg Walking those data structures for every access is going to be slow whether it's done in hardware or software. Caching would have helped. There's a detailed paper on why the 432 processor was so slow: |
The 432 was a strange system. Everything was an object, implemented in hardware with access control. You couldn't just move I/O data to memory because that would bypass the object protection. The Interface Processor translated between the object world and the I/O byte world.