Five Thousand a Second
Columbia's old page "Programming the ENIAC" was on Hacker News today. It calls ENIAC "about 300 times faster than the Mark 1 at addition". Here the two machines race, adding one each time, at the rates Wikipedia gives for each. You can slow time down until ENIAC's ring counters are slow enough to watch.
Each column is one decimal digit of a ten-digit accumulator, drawn as a ten-position ring counter with the lit lamp showing the digit. Units are on the right. ENIAC added by counting pulses round these rings and sent a carry pulse to the next ring whenever one wrapped from 9 to 0. Wikipedia says each digit took 36 vacuum tubes. The lamps here are a drawing of that logic, not of ENIAC's front panels.
The ratio
| Operation | ENIAC | Harvard Mark I | ENIAC faster by |
|---|---|---|---|
| Addition | 200 µs (5,000 a second) | 1/3 s (3 a second) | about 1,700× |
| Multiplication, ten digits by ten | 2.8 ms | 6 s | about 2,100× |
| Division | up to 28.6 ms | 15.3 s | about 535× |
On Wikipedia's numbers ENIAC was nearer 1,700 times faster at addition than 300. The Mark I worked on 23-digit numbers and ENIAC on 10, so the race is not perfectly fair to the Mark I. Even per digit, though, ENIAC comes out several hundred times faster. The same page calls ENIAC "150 feet wide"; Wikipedia gives about 100 feet long, 3 feet deep and 10 feet tall.
What stands on what
| Item | Source |
|---|---|
| ENIAC: 20 ten-digit accumulators, 5,000 additions a second, 200 µs cycle, multiply and divide times, ring counters, 36 tubes a digit, size | Wikipedia, ENIAC (citing Burks 1947, Wilkes 1956 and others) |
| Mark I: three additions a second, 6 s to multiply, 15.3 s to divide, 23-digit numbers | Wikipedia, Harvard Mark I (citing Campbell 1999) |
| "about 300 times faster", "150 feet wide" | Frank da Cruz, "Programming the ENIAC", Columbia University Computing History (2011, updated 2023), read via the Internet Archive because the live page sat behind a browser check |
| Ratios | My arithmetic |