Ah, the wonderful Atlas! (1960s British supercomputer)

[pier4r]

(03-24-2023 02:31 AM)Garth Wilson Wrote:  I sometimes think about how much money and effort was spent on computers that were, within even a couple of years, surpassed in power, affordability, and other desirable traits. 

This happens constantly nowadays too. Anyway one should not discount the development, especially in the past.

Nowadays one may have - more or less - the ability to port software from one mainframe/supercomputer to another (it is not strictly true either). In the past wasn't that easy.

In calculator terms, writing a program for a TI wouldn't easily port to the HP (from the 70s and 80s - and even nowadays). Therefore it is not only about the HW, rather it is also the development platform. It takes time to get to speed and changing HW too frequently would make the SW development stuttering.

All this to say, it is completely fine that systems, apparently obsolete in HW, get used for some more years because the SW development on another system would slow down again. On one side one gains HW speed, on the other one loses SW development speed. SW development speed is the larger part of the problem unless the product is a library used a gazillion times (in that case then the program speed is more important).

Just as a note. Sierra and Summit are still running despite (a) being installed in 2018 - their base HW is somewhat dated compared to what is available nowadays; (b) IBM not really pushing for major supercomputers at the moment.