Cray

The other day I read that “supercomputer” company Cray is being acquired by Hewlett-Packard. For a measly 1.3 billion dollari. Wasn’t WhatsApp purchased for 18 billion by Facebook? And Uber went public the other day with a valuation of over 80 billion? Without ever having made a profit. It’s a shame that a company such as Cray, which actually makes stuff, and is a legend, goes to HP for such comparably little money.

Their first machine, the Cray-1, was a 64-bit computer clocked at – wait for it – 80 Megahertz. Yes, Mega. With 8 MBytes of RAM. These were the days. Nonetheless, it was the fastest computer on the block. Seymour Cray, the founder of the company back in the 1970-ies, had realised early that the path to high computing performance is not just the CPU. He quipped: “Anyone can build a fast CPU. The trick is to build a fast system.” So he meticulously crafted each component and, even more importantly, their data and signal connections to a high precision, so that there were no lost computing and signalling cycles anywhere in the system. Mr Cray was a real engineer at heart. He made parallel signal paths exactly the same length, physically. He knew about controlling the heat in the circuits. He applied the Maxwell Equations to control the influence of high radio frequency effects. And high they were for this time, both regarding available technology and engineering knowledge.

My first Mac was running at 16 MHz, if memory serves. That was more than a dozen years after the Cray-1. I can remember very well – and this is embarassing to admit from today’s perspective –, when the computers’ clock frequencies started to raise, to 48, then just over 100 MHz, I was so sure that there would soon be a hard limit. I mean, 100 MHz is the wave transmission frequency of FM radio. And I had experienced myself what havoc high frequencies can wreak on my electronics boards, both analogue and digital. You get signal reflections and other nasty effects if your signal paths are not painstakingly laid out, and terminated properly. The electric signals that are supposed to travel along their metal paths on the board – well, they don’t anymore, but are all over the place. Bad Things™ happen.

A lot has changed since then. I was proven wrong, of course. Better technology, sophisticated knowledge, tools. Today, even your smartphone clocks at over 1 GHz (1,000 MHz). My PC runs at 4.5 GHz. To compare, your WiFi network’s radio waves oscillate at 2.5 - 5 GHz. Even though it’s interesting that the main clock on the computer’s main board is a mere 100 MHz, which then gets multiplied by the CPU (or some ancillary circuit) by 45. So it’s only well defined parts of the overall design, mainly inside the chips, and between closely positioned chips, that clock so high. Clever engineering can do a lot, but there are physical boundaries, especially if the cost have to be low. But still, we’re worlds apart from the Old Days.

Speaking of Old Days, I came across this article in my RSS feed.1 The meter presented there is really old school, but exactly what the state-of-the-art tech looked like and how it worked these days. Check out the metal shielding everywhere to diminish the electromagnetic interference between the components. My favourite element is the “oven controlled” reference oscillator. Oven controlled! And look at the size of the single boards and components, as well as their low integration! With this kind of technology, your portable phone would be the size of a shoe box, at least. Hence, not so portable. Schleppable maybe.

Remember when a 56.8 kBaud modem was state of the art? Probably not, but anyway, today, it’s possible to transmit several Mbit/s over the same old copper telephone wires. A 100-fold increase. My MSc degree is in Network Theory,2 but I have only faint ideas how this is even possible. Fast control of signal reflections and tight data encoding, I guess – I haven’t worked in this field since my studies. Life tends to interfere.

Back to Mr Cray, a real original. He preferred to work with paper and pencil. Or an electronic equivalent: when he was told that Apple Computer had just bought a Cray computer to help design the next Apple Macintosh, Cray commented that he had just bought a Macintosh to design the next Cray.


  1. Via Dangerous Prototypes. Nice moniker there. ↩︎

  2. And Biomedical Engineering. ↩︎