As part two (see previous attempt) of my ongoing series in `computational necromancy,' I've spent the last year and a half or so constructing my own 1/10-scale, binary-compatible, cycle-accurate Cray-1. [...] The Cray-1 is one of those iconic machines that just makes you say "Now that's a super computer!" Sure, your iPhone is 10X faster, and it's completely useless to own one, but admit it . . you really want one, don't you? [...]
When I started building this, I thought “Oh, I’ll just swing by the ol’ Internet and find some groovy 70’s-era software to run on it.” It turns out I was wrong. One of the sad things about pre-internet machines (especially ones that were primarily purchased by 3-letter Government agencies) is that practically no software exists for them.
After searching the internet exhaustively, I contacted the Computer History Museum and they didn’t have any either. They also informed me that apparently SGI destroyed Cray’s old software archives before spinning them off again in the late 90’s. I filed a couple of FOIA requests with scary government agencies that also came up dry.
Homebrew Cray-1A: Computational Necromancy
This is fantastic:
Tags: computers, mad science, retrocomputing, the future
I can't tell if I'm (a) amazed by the fact you can have a 1/10 scale Cray, duplicated clock for clock on generic programmable logic, on a hobbyist's budget or (b) disappointed that after 34 years of technical progress it's only half as fast.
I was amazed to learn that the Cray 1 was only about 200k gates. I think our perception of what supercomputers were is overblown. They were amazing for their time, but we've really surpassed them quite substantially with both modern microprocessors and modern supercomputers.
Keep in mind that what constitutes a "microcomputer", "minicomputer", "mainframe" or "supercomputer" *changes* over time.
I trained on a 60s vintage mainframe in the early 70s. I don't know how much RAM it had, but the manual for it (which is sitting on the bookshelf not 6 feet away) says that the *max* RAM was 32k 6-bit words.
These days a mainframe is *way* beyond that. :-). Likewise current supercomputers outclass the Cray by quite a bit.
I've been kicking myself for years about not taking up the offer a guy made back in the late 80s. He was *giving away* the "case" from a Cray.
It'd have made a nice centerpiece in the living room.
Mainframes haven't come that far. IBM still sells them by the MIP. ;)
These days mainframes seem to be more about supporting massive numbers of "terminals". And huge databases.
I'd go further: these days mainframes seem to be about padding IBM's bottom line. :)
Super computers in 1989 were about 10GFLOPS, which is more than a Core Solo. 20 GFLOPS computers hit the market in 1990, which is about on par with a 2.4 ghz Core 2 Duo... we aren't that far ahead of the late 80's.
A friend of mine is an admin at the Lawrence Berkeley Lab, and he worked on the old Crays. I've contacted him about the project. He was excited to hear about it.
I've decided to launch a more practical project and build a bench around my Mac Mini.
The way the vector-add opcodes are implemented remind me a lot of the kind of parallel implementations you can do now in CUDA on the GPU. On the shoulders of giants, and all that.
Yes, CUDA is a SIMD processing model, used by the Cray-1 and CM-1. These days, GPUs are pretty much what people thought of as supercomputers through the early 90s. Thus my crack about "Oh, you just wanted to place a light in the scene?"
Of course I want one. I have wanted one since CompSci lectures in the early 80s on them. The bottleneck in the original was apparently a pair of NAND gates...