FGCS'92: the ten years of the Fifth Generation summed up
On 1 June 1992 the last FGCS conference opened in Tokyo, where ICOT presented the results of the eleven-year project. By Kurozumi's overview the project cost about 54 billion yen. The 1982 goal was a prototype of a thousand processors at 100M to 1G logical inferences a second; at the start of March 1992 a PIM/m of 256 processors had just begun running tests, at a peak of more than 100 MLIPS. Fuchi says «as much as or more than what I expected» was achieved.
Why it matters
Five state programmes had beginnings without ends: this document is the end of one of them, written by the programme itself. It separates the goal (a thousand processors) from what was reached (256, and speed in applications that «may be much reduced»), and the decision to continue with a smaller team was announced in the same June. This is an editorial assessment.
The goal by the 1982 plan, as the report gives it: a prototype of a thousand element processors at 100M to 1G LIPS. Stages: initial 1982-84, intermediate 1985-88, final 1989-92; the final stage was planned for three years and after a halfway review extended to four, so the project ran eleven years. Budget: about 8 billion yen for the initial stage, about 22 for the intermediate, about 21 for 1989-91, 3.6 for 1992 (estimate), about 54 in all. Uchida: at the start of March 1992 a PIM/m of 256 processors «has just started to run a couple of benchmarking programs»; a PIM/p of 512 and a PIM/m of 384 processors are planned; for practical applications the speed «may be much reduced» from the peak. Fuchi: it takes another five years for the results to take root in society. Afterwards: by ICOT's 1994 report the decision on a follow-on project was made and announced at FGCS'92 in June 1992; it started in April 1993 and was to finish in March 1995; ICOT's budget by fiscal year, billion yen: 7.2 (1991), 3.6 (1992), 1.3 (1993), 1.3 (1994); ICOT researchers 90, 60, 40, 40. Since August 1992 more than 2,100 people have transferred more than 12,000 files of ICOT free software. That report was written two years later by the project's own division chief. What the record does not claim. This is ICOT's own assessment of its own project (rule 8): no independent evaluation was read, hence medium confidence. That the project «failed» or «succeeded» — ICOT's documents do not say so in those words, nor does the record. The number of PIM/p processors and the speed on applications beyond those given were not established.