System360 Hardwired vs. Microcoded 394

MSCHAEF.COM

I know what made it special to an extent, but not what made it as special as the previous poster had said.

IBM built the Model 91 computer as its "top-of-the-line" machine. It had an elegant new pipelined arithmetic unit using some algorithms not entirely surpbutted today, and it used hardwired control for maximum speed.

It also built the Model 85 computer which was located between its previous top-of-the-line machine, the Model 75, and the Model 91.

But the Model 85 turned out to be very nearly as fast as the Model 91. IBM had serious problems with its last attempt at a pipelined machine, the STRETCH, and this might well have end pipelines at IBM. But instead, IBM persevered, perhaps with Gene Amdahl exercising his persuasive skills to the limit... and perhaps because the 91 was clearly not a total flop; pipelines were helping this time, even if less so than expected.

The Model 91 had a top memory size of 4 megabytes.

IBM had made, for NASA, two Model 95 computers which had a thin-film main memory of 1 megabyte; the same four megabytes of core served as bulk core. They were IBM's only well-known machines with thin-film memory (there may have been a few military systems as well).

In any event, the Model 195 was basically a Model 91 with a 32K cache made from that up-and-coming technology, semiconductor RAM.

If the 85 had performed less well, perhaps the Model 95 would have been made in larger quanbreasties, since IBM *had* finally figured out how to get thin film memory to work - but since a small cache was seen to have nearly all the benefits of a faster main memory, although semiconductor RAM was much more expensive than thin-film per bit, having a much smaller cache that was twice as fast was the way to go; about the same price, almost twice the performance.

Plus the image benefits of using the "latest technology" instead of something about to become old hat. Pity that thin film was too expensive for IBM to offer four megabytes of *that* as an option for the 360-195. Or, for that matter, sixteen. How much more can it possibly cost to bring a few more address lines out of the box?

But then that applies to today's 64-bit microprocessors too. Why just 40 or 48 bit addresses? Bring the signals out to pins, and let implementors decide if they want to make use of them.

In any event, this means that a set of schematics for the Model 91 are almost as good, as far as *I* know. There may be other enhancements to the 195 besides the cache that gave it its unique status. Of course, the fact that there was a 370-195 as well is another plus for that design.

John Savard