Wanted: iAPX-432

At 09:25 AM 5/23/01 -0600, emanuel stiebler wrote: No. Maybe you are thinking about the i960? I dunno about that one, but it was a lot more CISC than the i860. BTW, intel is coming out with a net chipset that is called 860, but it is completely unrelated to the i860 processor. Hey, now. I've actually used an i860. A 3D graphics terminal I worked on in 91-93 used two: one for the control processor, one for the geometry unit (actually, we had the option of using two for the geometry unit, for a total of 3). For its time, it was the best deal around for doing floating point. Getting the maximum rated performance out of it required a lot of very careful hand scheduling of instructions, but that's true of most processors. I also believe it was the first microprocessor that had more than 1M transistors, though a lot of that was contained in the two caches. It could dispatch two floating point ops per cycle, and that was back in 1991. The 860 had a few problems. The big one was that intel didn't know how to position it. The second problem, which was slightly improved in a second version of the chip (that ran at 50 MHz), was that recovering from exceptions could be a bear. This hadn't really been thought through. I didn't work on it, but I recall hearing that the reference routine provided by intel ran a number of pages, and had to do things like check to see what the next instruction was and recover differently depending on what it was about to step into. In the worst case, it could take thousands of cycles to get back to a running state. I agree, intel has one butt-ugly architecture with the x86, but nobody in their right minds can fault them for succeeding despite that. Given the fact that they must live with the ugliness, their implementations are first rate. ----- Jim Battle == frustum(a)pacbell.net