Jump to content
  • entries
  • comments
  • views




Even though I didn't get into computers until several years after the Cray-1 first came to market, it was still a legendary computer.


Just the look and shape of it was awe inspiring; looking like something from 2001: A Space Odyssey (or any other SF film). No simple box, but a C shaped tower with a "love seat". Then you learned that "love seat" was the refrigeration system - it was cooled with Freon - how cool was that? A computer that got so hot it needed a cooling system. (Which actually wasn't that uncommon for large systems of the time. But when your point of reference is a fanless home computer....)


The actual design of the Cray-1 is interesting. Seymour Cray took the knowledge he gained working on the CDC supercomputers and built something very advanced.


One major design change he made was adding a huge number of registers for temporary storage, including vectors. Up to that point vector processors processed data stored in main memory. Seymour realized memory was too slow so a significant performance improvement could be gained by having a large number of registers and performing all operations on those registers. This is the same "load/store architecture" most RISC processors use. The Cray-1 had over 5 kilobytes of register storage!


The Cray-1 also had 12 independent functional units. Each was pipelined so they could process multiple instructions without delays. The vector and floating point units also had the ability to automatically forward the result to another functional unit, called chaining. With judicious use of this with vector processing the Cray-1 could hit 250MFLOPS under ideal conditions. (Although I'm not certain how, since it only had 3 floating point functional units and ran at 80MHz, thus 240MFLOPS.)


Yep, the Cray-1 had a 80MHz system clock; which doesn't sound that fast today. However, the Apple 2, available at the same time had a 1MHz system clock and the original IBM PC, which came 5 years later, had a 4.77MHz clock. So for the time 80MHz was insanely fast. Also consider the Cray-1 was made from LSI chips: 1Kbit SRAM for main memory (max 8Mbyte+1Mbyte ECC), 16x4 bit 6ns registers, and fast & slow NAND chips.


But while a microprocessor is typically less than an inch square, the Cray-1 "processor" was spread over 340 square feet circuit board (512 6x8 PCBs). So the design of the Cray-1 had to account for propagation delays, and ensure the delays were equalized as much as possible. (Either by cutting wires to precise lengths or adding extra capacitance to PCB traces.) The curve of the physical frame was also to minimize distance.


In order to be as fast as possible, the Cray-1 was made using ECL (emitter coupled logic) circuits. ECL circuits are very fast, but require a lot of power because they never "turn off". This led to the built-in cooling system, which always sounded so impressive.


Part of me wonders how small and how fast a Cray-1 made with modern technologies would be. There are certainly Cray emulators out there. But they probably only worry about duplicating the ISA, and are not clock accurate. Unfortunately, even if schematics were available it wouldn't be possible to simply "port" the design to modern technologies as much of the Cray-1's design was overcoming physical limitations. But it's fun to dream.


Recommended Comments

They have a real Cray-1 in the Deutsches Museum in Munich. The computer section is one of my favorite places in there.


I totally love going there. The most impressive exhibited object is an orignal V2 rocket though, which was assembled in the centre of a spiral staircase inside the building :)


BTW: If you ever come to Germany, make sure to visit the Europapark in Rust. They have the Russian "Mir" Space Station there. It's not just a replica, but the second original unit, which was used for crew training on earth! :)

Share this comment

Link to comment

There is a Cray-1 on display a few miles from where I live. It's in the lobby of Cray's new downtown St Paul, MN office. I've been wanting to go down there and take a look at it. It's a great piece of computer history.

Share this comment

Link to comment

There's a Cray in the movie Tron, where they're sneaking into Encom...





You can see part of the bench, too:





But... who's that guy? :)



Share this comment

Link to comment

IIRC TRON was shot at Lawrence Livermore Labs, who had one of the first Crays. The guy in the show is probably a grip or a best boy (I can never keep those two jobs straight). Or maybe a gaffer.

Share this comment

Link to comment

I've been inside LLNL, setting up simulation software on their NUMA SGI machines. Most of the scientists use OS X, running an X server to hit their big iron. Damn cool. A lot of the work there involves big memory model simulation, of the single image type, not the movie rendering type.


Anyway, CRAY was purchased by SGI. Part of that purchase was to get access to the CRAY patent portfolio and interconnect know how, and that's when the "NUMALink" was born, and SGI started down the road to NUMA (non-uniform memory access) multi-CPU, single OS image, modular computing, scaling way above the usual 16, 32 or 64 CPU systems, seen coming out of SUN and HP at the time. Memory intercommunication latency off the charts too, even up at 256 and 500 CPUs! That latency determines the scaling possible, and the productivity of the various nodes.


I've been on some 32 and 64 CPU NUMA, ONYX machines, running at a brisk 600Mhz, or 400Mhz, Mips R12, with powerful visualization graphics pipes, featuring hardware compositing and edge blending on multi-screen displays. Did some big scale CAD design, like modeling a chip fab, or huge building, and some video / animation stuff, like MAYA or Alias Studio for rendering and product design.


SGI transitioned off IRIX + MIPS, which could handle 1024 CPU, single system OS image, NUMA designs very well. A 2048 CPU was made for NASA, but I don't know that it ever ended up as one OS image... Many kinds of simulation need the fast memory interconnects, unlike simple frame rendering, which can be done Beowolf style, slow connects, multi OS image. (each node runs it's OS, communicating results to master nodes) NUMA machines run one OS, intercommunicating between processes and threads across the NUMA links, at insane speeds. Processors have 8GB, 4GB, 2GB memory local to them, fetching from whatever the system total RAM is, when needed. CRAY tech is a big part of that.


Anyway, they went to Itanium + Linux, porting over the IRIX scheduler, and memory interconnect software. Those systems ended up scaling to 1024, with several 512 CPU machines being common.


One other thing that came from CRAY was advanced compilers. All of those complex conditions, as well as the Itanium are extremely difficult to optimize. CRAY tech contributed there too, though not in as significant of a way. The high end computing found in NASA, NSA (scary data mining and visualization none of them wants to talk about), Petrofuel, medical life sciences, aerospace and auto all use some variations on the great stuff CRAY did, with most big iron Unix companies way behind where SGI + CRAY got to on single OS image, big memory, massive parallel computing.


Older school IRIX 8+ CPU systems can still fetch over 10K, running big ram, clocked below 1Ghz, still punching well above their weight over other systems on parallel computing problems that require the memory throughput. Until a few years ago, I was active in selling and supporting those boxes. Really fun hardware! The docs that ship on the machines are a complete comp-sci class all on their own.


CRAY contributed to big scale computing with their tech. SGI expanded on it, with the simple idea that all compute problems boil down to I/O problems, given the right memory interconnect technology, and a damn good compiler... Was at a convention in Vegas, one of those kick off kinds of things, where I got to talk to the SGI chief scientist and CEO for a coupla hours... It's flat out amazing the scale those guys think on!


CRAY was there at first, showing how to scale up really big, well ahead of the others.

Share this comment

Link to comment
Add a comment...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

  • Recently Browsing   0 members

    No registered users viewing this page.

  • Create New...