My uncle was one of the founders. My dad had a lisp machine in his home office for years. I never got to use it much, but I certainly spent a lot of time programming in LISP as a teenager.
I should see if my dad still has it, curious if it would still work?
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.
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).
This machine aged fast. I had one in 1986 https://youtu.be/P96Lp3MAJJ4?t=9m for personal use and I never really understood the purpose of this machine. It was not particularly fast compared to lisps in other machines.
If you compared the 3640 using low-level Lisp benchmarks, it was comparable to other similar expensive machines at that time.
The difference was mainly that it allowed you to write large Lisp programs inside a large Lisp IDE on top of a Lisp operating system. Unix-based Common Lisp IDEs were just appearing or about to appear in 1986 and large Lisps under Unix was still a challenge - for example when using GC and a large amount of virtual memory.
Your 3640 was basically second generation hardware. First were the CADR and then the initial 36 range. Later generations were slightly faster, supported newer hardware (SCSI, NuBUS, VMEbus, ...) and could be embedded into other machines (like the UX boards for SUNs or the MacIvory boards for Macs).
My uncle was one of the founders. My dad had a lisp machine in his home office for years. I never got to use it much, but I certainly spent a lot of time programming in LISP as a teenager.
I should see if my dad still has it, curious if it would still work?
Correction to the date in the title: this is a concatenation of blog posts from 2007 (not 2017). Specifically, these three posts:
https://web.archive.org/web/20120909104608/http://danweinreb... (Sunday, November 11th, 2007)
https://web.archive.org/web/20120909104443/http://danweinreb... (Friday, November 16th, 2007)
http://web.archive.org/web/20120909110022/http://danweinreb.... (Friday, December 21st, 2007)
There is other interesting stuff on that blog and in the comments etc.
Dan Weinreb died in 2012: https://en.wikipedia.org/w/index.php?title=Daniel_Weinreb&ol...
Thanks! We've updated the headline.
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.
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).
Makes me wonder if a CPU with an FPGA on board could use the fpga as some kind of "lisp microcode" interpreter.
The link in one of the posts to the Eve Phillips Masters’ thesis is now: https://www.ifis.uni-luebeck.de/~moeller/symbolics-info/ai-b...
Here's the “Symbolics, Inc: A failure of heterogeneous engineering” paper: https://www.ifis.uni-luebeck.de/~moeller/symbolics-info/Symb...
This machine aged fast. I had one in 1986 https://youtu.be/P96Lp3MAJJ4?t=9m for personal use and I never really understood the purpose of this machine. It was not particularly fast compared to lisps in other machines.
If you compared the 3640 using low-level Lisp benchmarks, it was comparable to other similar expensive machines at that time.
The difference was mainly that it allowed you to write large Lisp programs inside a large Lisp IDE on top of a Lisp operating system. Unix-based Common Lisp IDEs were just appearing or about to appear in 1986 and large Lisps under Unix was still a challenge - for example when using GC and a large amount of virtual memory.
Your 3640 was basically second generation hardware. First were the CADR and then the initial 36 range. Later generations were slightly faster, supported newer hardware (SCSI, NuBUS, VMEbus, ...) and could be embedded into other machines (like the UX boards for SUNs or the MacIvory boards for Macs).
For an interesting read from the MIT AI lab crowd check out this book:
http://www.blurb.com/b/2172660-minskys-trinskys-3rd-edition