Hacker News Reader

Comment by Animats | Hacker News Reader

Now if only someone can get this guy some Cray-I software. Even the Computer Museum, which has an actual Cray-I they use as a piece of furniture, apparently doesn't have any.

The Cray-I is a rather simple machine at the logic level. There are 64 of some registers, but they're all the same. The instruction set is small.

View on HN

To be fair, booting the Cray would require a good sized power substation. The power bill would be... impressive.

The 1s used between 100kW and 150kW. So a small substation. The XMP used 200-300kW. The 2 used 150-200kW.

There was a YMP model (the EL) that could be power off normal mains (220v).

Energy efficient was different back then.

I wonder if anyone's asked the National Cryptologic Museum/NSA

How did they code it, in assembly?

We used Fortran, although asm could be used and apparently there was a Pascal, but I never saw it. We actually did the dev cycle on a front end system (VAX/VMS or Unix) and submitted the program as a batch job (COS was a batch OS, the multiuser UniCOS didn't show up till much later).

Wow, thanks for replying.

I must have been an incredible opportunity working with a supercomputer. Did you ever compute a mathematical constant or an algorithm to a large precision out of curiosity?

Yeah...I got lucky. My third year or so at Georgia Tech a professor offered a "special topics" (e.g. not in the course catalog; his personal research interest) undergrad class in supercomputing. GaTech was good for little bonuses like this (I took Micheal Barnsley's first IFS class, too).

In addition to Crays, we got time on a CDC Cyber 180/990 and a 205, an ETA machine, and had cursory intro to NEC SX, BBN Butterfly, Multiflow Trace and a few other weird things noone has ever heard of. We wrote a handful of numerical methods programs for each (so, yes, computing mathematical constants and such, but more vector/array manipulations), and looked at how the architecture of the machines (short v. long vectors, register based v. memory based, vector v. MIMD v. VLIW) effected the speed of the programs and what optimizations/refactorings resulted in large speedups.

Good times.

Reply on news.ycombinator.com