The Cray 2 Super Computer (1985) [pdf] | Hacker News Reader

The Cray 2 Super Computer (1985) [pdf]

archive.computerhistory.org

42 pointsChuckMcM29 comments

I had the opportunity to use a Cray Y-MP 2E at university in the early 90s. At the time there were strict export restrictions, so I had to get a special approval from the U.S. Government (this was in Brazil).

It was a beautiful machine. The C-shaped purple machine was impressive, and the refrigeration system took most the space. And we had some nice SGI Indigo as front-end workstations, so you could optimize your (Forth) code before moving to UNICOS - the Unix flavor used by Cray.

In hindsight, it's ironic to think about all the paranoia and secrecy - for just 666 Megaflops. My not-so-new Core i7 has 150,000+ Mflops, not counting the GPU.

Ahh yes, modern computers put these older systems to shame in some regards, but what did the typical desktop have back then?

There's a person out there who made a 1/10th scale cycle-accurate Cray-1A...which is truly pretty cray http://www.chrisfenton.com/the-numbotron/

If we say that about the Cray 2 30 years later, interesting to think what they will say about the cloud in 30 years.

I'm always annoyed at the "my shiny new desktop is faster than that old Cray". A better comparison would be "how fast a computer can you buy for 2m 1990 dollars adjusted for inflation?". That's more than US$3.7m; that's a lot of computer.

I got to use a Y-MP48 in the old days. Good times.

I find it interesting to review and contrast the architecture of super computers with modern day processors. The clock speed and memory quantity are higher, but often there are large differences in memory bandwidth and I/O bandwidth. This shows the Cray 2 with multiple 4 Gbit paths to I/O (think PCIe x ~2).

This table indicates that PCIe v2 is 4 Gbit/s per lane (64 Gbit/s for 16 lanes) and PCIe v4 is almost 16 Gbit/s per lane (up to 252 Gbit/s total). Is this correct or am I misunderstanding what you are saying?

http://en.wikipedia.org/wiki/PCI_Express#History_and_revisio...

I was thinking PCI Express, Version 1, by 2 lanes. There are four of those if I read correctly which would be essentially 8 lanes of ver1 PCIe.

Rather a bit more I/O bandwidth than your iPhone has, but not as much as say a current desktop.

Here is an interesting page: http://en.wikipedia.org/wiki/List_of_device_bandwidths

"Whistlebower" Dr. Bill Deagle, MD claimed that the unique Gallium Arsenide chip technology that was being worked on by the Cray Computer Corporation essentially became classified by the US Air Force, and that it was developed much further than is publicly known, and that he witnessed a "Cray 5" computing array in operation. https://youtu.be/tOz0-Hy1rm8?t=462 (7:41)

When you compare the properties of Gallium Arsenide to Silicon chips, it makes a lot of sense that the military would be highly interested. http://en.wikipedia.org/wiki/Gallium_arsenide#Comparison_wit...

That guy is bogus. There have been some real cryogenic computing devices, though. IBM and NSA put huge efforts into that back in the 1950s. ("I want a thousand megacycle computer. I'll get you the money" - NSA director).[1] The first generation technology, cyrotrons, sort of worked, but mainstream technologies pulled ahead, and that technology was abandoned in 1965. IBM kept plugging away at cyrogenic computing through the 1980s, with Josephson junctions being the next technology. Those run at liquid helium temperatures. With great difficulty and at huge cost, they got some experimental electronics (not a full CPU) running at 300MHz. By this time, it was 1982, and while that was about 10x faster than the fastest microprocessors of the period, it looked like the upper limit of the technology was around 1GHz. That, plus high cost, plus all the headaches of working in liquid helium, indicated the Josephson junction was a dead end. Standard CMOS was going to outperform the exotic technology. So, in 1983, IBM pulled the plug on that project.

They're trying again.[2] IBM is working on superconducting quantum computing, and has been busy since 2012. Maybe this time it will be useful. It will definitely be expensive.

[1] http://www.ewh.ieee.org/tc/csc/europe/newsforum/pdf/RN28-1.p... [2] http://defensesystems.com/Articles/2014/12/04/IARPA-cryogeni...

Seymour Cray, the supercomputer architect, set up a new company, SRC Computers, and started the design of his own massively parallel machine. The new design concentrated on communications and memory performance, the bottleneck that hampered many parallel designs. Design had just started when Cray died suddenly as a result of a car accident.

He died on October 5, 1996 (aged 71) of head and neck injuries suffered on September 22, 1996 in a traffic collision. Another driver tried to pass Cray on Interstate 25 in Colorado Springs, Colorado but struck a third car that then struck Cray's Jeep Cherokee, causing it to roll three times. Cray underwent emergency surgery and remained in the hospital until his death two weeks later. The Jeep Cherokee was designed on a Cray super computer, what a sad coincident. (source "The Superman" book about him)

When he was told that Apple Computer had just bought a Cray to help design the next Apple Macintosh, Cray commented that he had just bought a Macintosh to design the next Cray.

http://en.wikipedia.org/wiki/Seymour_Cray#Cray_Computer_Corp...

Listening to him talk really doesn't inspire much confidence. Especially his claims that GaAs "enables quantum computing".

Well, sure. The Cray-5 is in the building next to the Crashed Alien Saucer Hanger at Area 51. Fitting, since Seymour came to Earth in one of them. Also, "whistleblower" is what you use instead of "complete crank" in polite company.

There is a lot of classified work on GaAs for communication links where the speed of demodulating/modulating a signal is critical.

But for general computing, implementation is more critical, consider that the Cray T3E sustained 1 teraflop of computing power, while 3 years ago the Cray-3 could sustain only a meager 16 gigaflops - clearly silicon could deliver performance (and it does so today!).

To put the specs in more modern terms, it's roughly a 64-bit quadcore with 2GB of RAM and a clock speed around 250MHz. Enormously powerful for 1985, but probably slower than a typical desktop system in 2005.

No mention of secondary storage... but with 2GB of RAM, maybe it didn't need any?

Disk channels to SMD or SCSI disk subsystems (maybe IPI but those disks were temperamental at best).

As I recall it wasn't until SSE3 instructions which were in the mid-2000's that an x86-64 desktop could compute an image convolution in the same time it took the Cray 2 but I'd have to ask Ram Nevatia. He was always finding the fastest way to do various image analysis programs.

A few years ago, somebody got a Cray 2, barrels of Fluorinert and all, and was going to restore it. Whatever happened to that? Probably not much.

Restoring old computers is a huge task. The Computer Museum in Mountain View worked for years to restore an IBM 1401, a small mass-produced machine from the 1960s. They had techs who'd worked on them and access to the original designers, and it was still very hard.

Kids these days with their data center-sized supercomputers which are a bunch of x86 rack servers in a big cluster. Back in the 80s and 90s, supercomputers looked the part, and had cooling waterfalls

The biggest challenge may be powering it up. The brochure says it sucks down 300 kW. I'm guessing it'll need a 440V polyphase feed to run the 400 Hz motor generators...

Yeah...the Cray 2 basically required it's own power substation be installed.

I started to read this and remember seeing [this][1] from a couple of years back... with the liquid cooling, etc, i would love a replica of the 2...

[1]: http://www.extremetech.com/extreme/191877-how-to-build-your-...

Ah, heady days when the world's best computers were made in Chippewa Falls, Wisconsin and featured waterfalls of Fluorinert for cooling.

By the time I saw a Cray-2 in person, it was shut down (it was so outmoded it was not worth running). But an awesome sight.

As a Wisconsin native, that Cray was situated in Chippewa Falls was always a fun piece of tech trivia.

So 2GB of memory, 4 cores at ~250Mhz (was the foreground processor the same? Brochure seemed unclear), runs a *nix. About 15 years ahead of a late 90's Pentium I guess, but more ram. Just need the spare 16 square feet of floor space.

and 300KW of power to run it!

And don't forget that you also have to get rid of 300kW of heat. Having worked in a supercomputing center, this is a major problem — think about how quickly you have to shut your computers down if your heat exchanger goes down, before the room gets too hot to enter.

Roomer has it that the original Cray 1 design didn't have a tunable cooling unit. One day the test machine crashed, and the engineers came in the next morning to see a giant icy glacier.

100 domestic fan heaters running at the same time? Toasty...

It's not that the Cray 2 isn't amazing, it's that modern chip tech puts at least that much power in your pocket, while desktop chips are far faster.

Aaaaand - it's all air cooled and apparently effortless, and the CPU does it all with <100W or so, and it's so cheap almost anyone can buy it.

Really, that's at least as amazing as the Cray series.

Yes, but can it run Crysis?

Reply on news.ycombinator.com