On Wed, Jul 26, 2017 at 1:50 PM, Ed Sharpe via cctalk <cctalk@classiccmp.org > wrote:
> define hardware diff. 635 and 645......ed# > > Disclaimer: I wasn't there; this is my understanding from my work on the emulator. The 635 was GE's flagship mainframe. 36 bit word, 256K word limit on memory. No virtual addressing, but 'base and bound' addressing. (A contiguous region of memory contain "user" program would be set up to appear to it as it was occupying memory starting at address 0. Physically, the memory would contain the operating system in low memory and user programs in high memory. When the OS task switched to the user program, the 'base and bound' memory location made the user program appear to start at address 0 and the operating system memory would be hidden (and protected) from it. The OS was GECOS (or GCOS), I forget which came first). The Multics project selected the 635 as the base of a new machine and worked with GE to develop the 645 to meet the needs of Multics; most specifcally the "Append Unit" (APU) which implemented Multics' [unusual] virtual memory model. Functionally, the APU sits between the CPU and memory and maps virtual addresses to physical ones. In reality, it invades the entire CPU; new registers, new instructions, new addressing modes, the whole concept of page faults, extending the maximum memory configuration from 256K words to 16M words. The 635 had two modes, absolute and BAR (base and bound). The 645 had a new mode, "append" (and much to the consternation of the emulator developers, also 'append with base and bound'; Multics uses that mode to run GCOS programs in a GCOS emulator). In absolute and BAR modes, the 645 is quite similar to the 635, but once you switch to append mode, it gets pretty wild. At some point in the evolution from the 635 to the 6180, the instruction set was extended with string and decimal math operations: Extended Instruction Set (EIS). I don't know the timeline or model designations. -- Charles
