Cray-2 Memories [pdf] | Hacker News Reader

Cray-2 Memories [pdf]

0x07bell.net

88 pointspinewurst20 comments

I know a guy who had a Cray at work (at one of the spooky national research labs that will remain nameless), which was of course on display in a room with big windows that visitors could look through to admire it. He knew a particular tour group was coming through, so he hid inside the Cray until they arrived at the window. Then he walked out of the Cray while zipping up his fly, looked up and saw the people gawking at him, pretended to be startled, flashed an embarrassed grin, and shuffled away.

All of the labs I’ve been to, including ones doing “weapons development” (shorthand for nuclear weapons programs, at least amongst my colleagues, who were not connected to those programs directly), have been very much like other areas of academia. Significantly more security, of course, and there are restricted areas, but generally quite nice, and I don’t think anyone who is used to working in a research lab or research university would feel out of place.

Far more spooky is the lack of amenities. I did my PhD in a program where I worked some of the time at Brookhaven National Lab, and both the dorm and cafeteria there were just embarrassing. It honestly was one of the reasons I left academia after awhile. It was hard to feel appreciated when your tiny dorm rooms frequently had broken AC/heat/refrigerators, and you shared your space with (impressively large) bugs.

Dorm at Stony Brook? Or does BNL actually have a dorm?

I bet BNL was just like this video: https://youtu.be/2310P2g1z2E

BNL does have dorms, and there is a reason none of them made it into that video, haha.

Seeing all the shots in that video made me quite nostalgic, thanks for sharing. I’ve never seen that before :D

The DOE labs weren't so spooky. Lawrence Livermore was more like a campus with public bikes at the end of the 8O's. I got to program on the Cray 2 in FORTRAN as part of a DOE high school program. I even got a picture of myself standing by the Cray 2. They did take security seriously as it's the only time I've had a M-16 pointed at me.

Aren't you, uhh, not supposed to aim at targets you don't intend to fire on?

Well, yes.

Apparently, at the time, bringing a foreign national (such as an European exchange student participating in the same program to see the host individual's non-classified experiment) to the lab after 5PM was a bad thing. Bad things at national labs get an armed response very, very quickly. Armed responses are by definition pointing weapons at people who they might need to shoot.

It was actually more surreal than scary for me, but I had grown up on a rez and was a bit more fatalistic than the other passengers in the car. I guess we made for a rather harmless test of the response speeds.

Well see, that’s the thing. They do intend to shoot you, but only if you don’t do what they want.

No idea what the context here is, but I’d assume military (adjacent? Do uniformed servicemen protect DoD sites and co?) security wouldn’t hesitate to point live weapons at you if you have to prevent someone from <x>. Unless you’re ready to immediately fire, you might not stop them from <x>.

That’s how I reason, anyway. No personal experience thankfully ;-)

Nuclear facilities of any sort are a good example of places where guns can get pointed at you, even if you're a U.S. Marshal on your way in to apprehend a fugitive.

https://www.funraniumlabs.com/2018/09/the-fringes-of-regulat...

Well he shouldn't have used tabs instead of spaces, then!

Now I want a restroom done in Cray decor.

I got to see the Cray Monsanto Corp owned when my grandpa retired (he was a senior engineering fellow in their old petroleum division, so he had some pull).

I can't remember specifically if it was the Cray-2, but it was definitely one of the models with that shape of coolant fountain.

Another crazy thing he showed me was the labs where they were developing the genetically engineered pesticide resistant grain strains. They were longish rectangular tunnel shaped concrete rooms about half below ground. All light was artificial.

The last big thing I remember was an autosteroscopic display, sort of like those lenticular illusion card things you'd get as a kid in that era on steroids. He said they used it for CG visualizations of molecules in his dept.

I was just 11 at the time so I was totally ignorant of the significance and controversy those plants represented. But I do remember grandpa pointing out the building the CEO and related staff worked out of. It was literally a bunker separated from everything else by big berms and security checkpoints that looked like it was securing a nuclear weapons depot.

It's not hard to emulate the early Cray machines. Someone did it in an FPGA.[1] They're not that complicated; they just have large numbers of identical units. But apparently almost all Cray software has been lost.

[1] http://www.chrisfenton.com/homebrew-cray-1a/

A big part of the performance of these machines came from high bandwidth memory access. This is high performance more in the sense of an 18-wheeler than in the sense of an F1 car. The point is more about predictable control flow and bulk manipulation of data than it is lots of tricky control flow, etc.

Of course, this analogy isn't really complete in that it misses the fact that Cray's machines also pushed the limits on cycle times. As his career evolved, his focus shifted from circuit design to reduction of propagation delay with more tightly packaged machines, then to cooling those tightly packaged machines, and then ultimately to GaAs as a way to get switching times down.

So not just an 18-wheeler, but one that happens to run at 200 mph, with all sorts of custom high-end engineering to make that happen with technologies of the era. The fact that these days, we can emulate this with programmable hardware is an amazing testament to exactly how much hardware has improved over the years.

Something else I should mention is that Cray also had an element of design conservatism to his work. The Cray 1 is famous for being a vector supercomputer, but it was not the first - his former employer, CDC, beat him to the punch with the CDC Star. Unfortunately for CDC, they got the balance of the design wrong. The Star wound up with vector capabilities that took too long to set up to be performant and scalar performance that was too slow to make up the gap. Cray's design for the Cray 1 got around both of these limitations, which is part of why it was as successful as it was.

The only time I've ever seen a Cray in person was in a barn, behind a friend's house.

At my first startup, there was an older engineer named Spence, who had a hobby trading used scientific equipment. His motto was, "He who dies with the most stuff wins". In the barn behind his house was a huge assortment of used equipment, including a Cray, a 9 million volt VandeGraaff, multiple CNC lathes, and a veritable sea of printers & monitors. His Cray was in bad shape and didn't run. The boards looked really interesting. He had hopes of getting it working again, but sadly, he got hit with a double whammy of cancer and divorce, and had to sell his barn and everything in it.

Working with Spence was a trip. I remember once when we were working on a vacuum problem and someone said, "Man, I wish we had a vacuum gauge that would measure down to 10^-9 torr". Spence reached into his pocket and pulled out a vacuum gauge, and said, "You mean like this?". We were never sure if he somehow anticipated what we would need, or if he was just the kind of guy who walked around with vacuum gauges in his pockets.

That PDF has very nice high resolution pictures I haven’t seen before.

At a previous job I was receiving training from a Cray engineer and things were winding down as we approached lunch. We were talking about various hardware replacement activities and the instructor recalled such work he had done on older Cray machines, I think it was a Cray-2. My memory is fuzzy around how exactly this came up because upon hearing "So I was replacing a register" my mind detonated. I was aware processors had spanned multiple chips and boards but hadn't really internalised how not long ago that was until that moment.

I remember way back in 1990 when I first got to my university, salivating at the idea of getting access to their Cray-2. It had been paid for by the DOE for a research lab, and basically nobody got access to it. It took me three years to work my way up the computing center ladder, but I was finally able to get an account on it through connections.

UNICOS was weird, and the interactive shell was slow as molasses. But it sure ran my raytracing jobs a heck of a lot faster than a Sparcstation 5!

That's because Cray's didn't have MMUs much less paging functionality. All the jobs required manual shuffling around and even a certain amount of explicit bounds checking.

Essentially they were batch Fortran machines, CDC6600+++ with larger memories with modern-ish OS hacked onto them.

Cray-2: nearly 2 GFlops. Apple Watch 1: around 3 GFlops. M1 MacBook Air: nearly 1 TFlops.

Reply on news.ycombinator.com