From: anton@mips.complang.tuwien.ac.at   
      
   scott@slp53.sl.home (Scott Lurndal) writes:   
   >One must also keep in mind that the VAX group was competing   
   >internally with the PDP-10 minicomputer.   
      
   This does not make the actual VAX more attractive relative to the   
   hypothetical RISC-VAX IMO.   
      
   >Fundamentally, 36-bit words ended up being a dead-end.   
      
   The reason why this once-common architectural style died out are:   
      
   * 18-bit addresses   
      
   * word addressing   
      
   Sure, one could add 36-bit byte addresses to such an architecture   
   (probably with 9-bit bytes to make it easy to deal with words), but it   
   would force a completely different ABI and API, so the legacy code   
   would still have no good upgrade path and would be limited to its   
   256KW address space no matter how much actual RAM there is available.   
   IBM decided to switch from this 36-bit legacy to the 32-bit   
   byte-addressed S/360 in the early 1960s (with support for their legacy   
   lines built into various S/360 implementations), DEC did so when they   
   introduced the VAX.   
      
   Concerning other manufacturers:   
      
    tells me that the   
   GE-600 series was also 36-bit.  It continued as Honeywell 6000 series   
   .  Honeywell   
   introduced the DPS-88 in 1982; the architecture is described as   
   supporting the usual 256KW, but apparently the DPS-88 could be bought   
   with up to 128MB; programming that probably was no fun.  Honeywell   
   later sold the NEC S1000 as DPS-90, which does not sound like the   
   Honeywell 6000 line was a growing business.  And that's the last I   
   read about the Honeywell 6000 line.   
      
   Univac sold the 1100/2200 series, and later Unisys continued to   
   support that in the Unisyst Clearpath systems.   
      
   says:   
      
   |In addition to the IX (1100/2200) CPUs [...], the architecture had   
   |Xeon [...] CPUs. Unisys' goal was to provide an orderly transition for   
   |their 1100/2200 customers to a more modern architecture.   
      
   So they continued to support it for a long time, but it's a legacy   
   thing, not a future-oriented architecture.   
      
   The Wikipedia article also mentions the Symbolics 3600 as 36-bit   
   machine, but that was quite different from the 36-bit architectures of   
   the 1950s and 1960s: The Symbolics 3600 has 28-bit addresses (the rest   
   apparently taken by tags) and its successor Ivory has 32-bit addresses   
   and a 40-bit word.  Here the reason for its demise was the AI winter   
   of the late 1980s and early 1990s.   
      
   DEC did the right thing when they decided to support VAX as *the*   
   future architecture, and the success of the VAX compared to the   
   Honeywell 6000 and Univac 1100/2200 series demonstrates this.   
      
   RISC-VAX would have been better than the PDP-10, for the same reasons:   
   32-bit addresses and byte addressing.  And in addition, the   
   performance advantage of RISC-VAX would have made the position of   
   RISC-VAX compared to PDP-10 even stronger.   
      
   - anton   
   --   
   'Anyone trying for "industrial quality" ISA should avoid undefined behavior.'   
     Mitch Alsup,    
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)