Los Alamos—High Speed Parallel Interface

(1976, age 29)

In the 1970s, the Los Alamos National Laboratory (LANL) in New Mexico used computers to simulate thermonuclear explosions rather than actually exploding a nuclear device underground at the Nevada test site. Simulating a nuclear explosion required significant computing power. Though Los Alamos had several CDC 6600 and CDC 7600 computers, that wasn’t enough.

I was a member of the technical staff in the C-9 group working at the Central Computing Facility (CCF) at Los Alamos in 1976. Cray Research had just announced the Cray-1 supercomputer and LANL got into a bidding war with Lawrence Livermore National Laboratory for serial number 001. We won that battle and in 1976 the most powerful computer in the world arrived at our computing facility.

8X02 internal diagram

There was just one problem. None of our other computers could “talk” to the Cray. The interface was different and the signal timings were very fast– faster than any of the other machines we had–faster and different than any other computers anyone had. No other machine could talk to the Cray, including the huge terastore system that would hold all the data generated by the Cray-1’s simulations. What we needed a computer interface box that could talk to the Cray at Cray speeds and connection protocols and could also talk to the other machines on the LANL network.

I was assigned to build that computer interface. I had to come up with a design and Weldon Scoggins, my technician, had to build it, by hand, on wire-wrap boards. There were just the two of us to make this all happen. Since we already had the Cray computer on site, this had to be done in a hurry.

I knew the logic in my design had to be blazing fast and I knew I had to use commercially available integrated circuit parts. There was no time to design new components. I needed something that was faster than the signals I would be getting from the Cray.

I chose to build a finite state machine around a very fast 8X02 Control Store Sequencer. This was an odd choice and I was pretty sure the 8X02 had never been used in the way I envisioned. The 8X02 is a simple but very fast address controller. This won’t mean anything to most people, but when I saw this diagram I knew immediately that I had everything I needed to tame the Cray-1.

8X02 internal diagram

Though it wasn’t a computer, the 8X02 did have instructions in a sense but they only determined the next address in the microcode that would be executed. The trick was to have additional bits at those addresses control, through hardware, some piece of the Cray-1 interface. Thankful for my previous experience with DEC computers, I wrote an assembler that ran on the PDP-11 to generate instructions to burn into bipolar PROMs for the HSPI. Those PROMs defined what the entire machine would do, so circuit changes weren’t made with wires–they were made with microcode revisions. This allowed rapid development and testing. Here is a picture of the first HSPI that was put into service at Los Alamos.

LANL HSPI

It struck me that I had built a finite state machine to talk to the world’s most powerful computer and that I had built a finite state machine before, when I was 14 years old, as a 9th grade science fair project. That earlier machine was built out of surplus telephone relays and the successive states were switched by a human pressing a button every few seconds. The HSPI state machine switched states every few billionth’s of a second. But the principle was exactly the same.

The HSPI was a success at Los Alamos. My design was kept simple to allow for maximum speed and scalability and to put it into service quickly. Our boss, Don Tolmie, went on to propose it as a standard for supercomputer communications. We were labelled as the “lunatic fringe–who in the world would need anything that fast?!” But we knew the computing world would move on to faster machines. The HSPI became the HIPPI (High Performance Parallel Interface) and it was popular for a decade to connect supercomputers to each other point-to-point and to attach high-speed storage devices.

I recently visited Los Alamos. I couldn’t get into the Central Computing Facility. My “Q” clearance has long expired. No matter, though, because the Cray-1 that I wanted to see isn’t in the CCF. It’s on display at the Los Alamos museum sitting silently, powered down forever, though it is still running smoothly in my mind.


HSPI side view, showing connectors.
8X02 Technical Reference
8X02(s), clock, PROM memory chips