I wrote:
Still working on the de-speckle problem. It turns out it de-speckled
away most of the dots on "i" and "j"s and several periods.
Computers are literal minded little devils!
Whatever happened to Asimov's intelligent computers that will solve
the energy crisis with a bold breakthrough in 2061? I feel cheated.
What gets me is that Asimov imagined that Multivac would not
understand plaintext English, but it would smart enough to invent new physics:
". . . Adell was just drunk enough to try, just sober enough to be
able to phrase the necessary symbols and operations into a question
which, in words, might have corresponded to this: Will mankind one
day without the net expenditure of energy be able to restore the sun
to its full youthfulness even after it had died of old age?"
Also, Multivac is made with transistors but for some reason it sorts
data "with contented lazy clickings . . ." (Later, the transistors
are replaced with "molecular valves" -- a great name.)
Computers sure are faster though. They have increased in speed by
roughly 11 orders of magnitude between 1945 and 2002. That's
according to this and database of Computer System Performance,
measured in MIPS on a log scale, which I found the other day. This is
interesting, with lots of information crammed into a small space,
which is the kind of thing I like:
<http://jcmit.com/cpu-perf-chart.htm>http://jcmit.com/cpu-perf-chart.htm
The database is here:
<http://jcmit.com/cpu-performance.htm>http://jcmit.com/cpu-performance.htm
This table is easier to deal with if you copy it into a spreadsheet.
It seems carefully researched to me, with painstaking conversions of
different benchmark programs into a common metric.
I quibble with his first and lowest data point, for the Harvard Mark
I, 1.0E-07 MIPS. I would start with ENIAC, just above 1.0E-05.
Notice the outlier point in 1955, back below 1.0E-05. That's the NCR
303. I do not know anything about it, but there were some other
computers of that era with simlilar slow performance that used
magnetic drums instead of RAM. They had "one word of RAM" as a
programmer of that era once told me. The IBM 650 was the prime
example. According to this table it was slower than the ENIAC. But it
was still a commercially successful machine, and very useful. This
goes to show that sometimes retrograde, slower, less capable
technology works better than the state of the art.
Notice that the gap between the slowest and fastest machines closed
in the 1990s early 2000s. (Except for the Intel ASCI Red outlier, up
at 1.0E+06.) That is my recollection. I expect that with things like
the Google-plex massively parallel computer this gap has opened up again.
- Jed
