4 ms·

Most special purpose machines failed because it took 2-5 years to engineer and produce each new generation of hardware. In the meantime generation purpose machi

Most special purpose machines failed because it took 2-5 years to engineer and produce each new generation of hardware. In the meantime generation purpose machines like Intel or Motorola CPUs came out every 6-12 months. An early 10x price/performance od special over general lead would evaporate next special iteration cycle. I saw this happen in LISP machines, massively parallel processors, array processors, etc. throughout the 1980s. My university lab acquired a FPS, MassPar, Connection Machine, Silicon Graphics only to see general purpose catch up. All the fancy hardware and firmware tricks to accelerate LISP could be emulated in faster in software after a few new CPU generations. Eventually a nimble company like Nvidia found a good mix of generallity and specificity to succeed. The semi-standard ARM specifications are another way the mobile computing world stayed in the semi-custom game.

lispm 8y ago
I don't think they failed. They had served well in a time window from early 70s to the end 80s or even 90. In the US much of the stuff was research and development, and a lot was paid by DARPA (the defense research agency). For example something like a LISP Machine was originally not developed as a fast Lisp system. It was in the mid 70s the dream of a computer for software development in Artificial Intelligence. The bottleneck was not so much processor speed, but memory and that computers were shared with people on time-sharing systems. 1 MB RAM and 50 users. The system speed was the problem - not the processor speed. The Lisp Machine was designed as a single-user workstation with GUI - in mid 70s. Since the researchers wanted them for Lisp and they mostly knew Lisp, they also wrote the OS and the design software (for example chip design) in Lisp. A commercialized GUI computer like Mac from 1984 was a tiny machine without development tools: floppy drives, 128k memory and a tiny screen. That stock hardware was eventually faster was no surprise, since a) Lisp processors were not particular fast and b) people did run simpler Lisp software on those - mostly applications, but not a Lisp OS, which was no longer needed. The Lisp processor made it faster to develop code. One of the main purposes was to be technology accelerators in the cold war of the 70s/80s - up to the Strategic Defense Initiative (SDI).
kristianp 8y ago
Makes me wonder if a CPU with an FPGA on board could use the fpga as some kind of "lisp microcode" interpreter.

lispm 8y ago

I don't think they failed. They had served well in a time window from early 70s to the end 80s or even 90. In the US much of the stuff was research and development, and a lot was paid by DARPA (the defense research agency). For example something like a LISP Machine was originally not developed as a fast Lisp system. It was in the mid 70s the dream of a computer for software development in Artificial Intelligence. The bottleneck was not so much processor speed, but memory and that computers were shared with people on time-sharing systems. 1 MB RAM and 50 users. The system speed was the problem - not the processor speed. The Lisp Machine was designed as a single-user workstation with GUI - in mid 70s. Since the researchers wanted them for Lisp and they mostly knew Lisp, they also wrote the OS and the design software (for example chip design) in Lisp. A commercialized GUI computer like Mac from 1984 was a tiny machine without development tools: floppy drives, 128k memory and a tiny screen. That stock hardware was eventually faster was no surprise, since a) Lisp processors were not particular fast and b) people did run simpler Lisp software on those - mostly applications, but not a Lisp OS, which was no longer needed. The Lisp processor made it faster to develop code. One of the main purposes was to be technology accelerators in the cold war of the 70s/80s - up to the Strategic Defense Initiative (SDI).

kristianp 8y ago

Makes me wonder if a CPU with an FPGA on board could use the fpga as some kind of "lisp microcode" interpreter.