The Seymour Cray Era of Supercomputers | Hacker News Reader

The Seymour Cray Era of Supercomputers

ztoz.blog

49 pointsingve13 comments

I was part of a group from the University of Minnesota who took a road trip to the Cray plant in Chippewa Falls, Wisconsin. I don't remember the year, but it was early - they were doing final tests on serial #5 of the Cray-1. My expectation prior to the trip is that the closest we would get to the computer was the other side of a glass window, but we got up close and personal. We were invited to touch the panels separating the inside and outside, they were very proud of their cooling system. One small trip into the inside could have caused millions of dollars of damage to the wire nest.

We briefly met the great man himself in his office. I don't remember much of what he said that day, except for the story of the pumpkin on his desk. His daughter had grown it, and knew he was building the Cray-1 and starting the design of the Cray-2. So she called the pumpkin the Cray-3.

I was around there quite a while after Cray himself passed away, but even then, people who were part of the OG Cray Research, even a single-digit handful that started at Control Data and followed all of the acquisitions that spent their life at that institution, still spoke about him with absolute reverence.

Until they moved offices, his portrait hung in one of the main conference rooms in the St Paul office and I could tell his ghost had a profound effect, to visitors and employees alike. I am supremely envious of you having been able to speak with him, was one of the few people in tech that I looked up to back then.

The original Chippewa facility (1050 Lowater) is something else. Room after room of systems in various states of completion, test fixtures, pallet after pallet of CPUs and ram, in the time of the massively-parallel systems, and all other parts of the machines themselves ready for assembly. Test equipment dotting every corner during bringup of parts of a new architecture. One room held customer engineering and replicas of legacy systems still chugging away in case a high-profile customer needed assistance. I believe it was an X1 reserved for a petroleum company. And yes, being able to walk around and actually touch a $150 million system running pre-delivery acceptance tests that would have been shipped out a few weeks later and installed at a customer site was unreal, not to mention having committed software that would have been running on that same system. Just a place like no other and I don't think it will ever be duplicated.

I still kick myself for leaving and wasting untold amounts of time at shitty, mind-numbing, software startups that went nowhere and had comparatively zero, inching toward negative, impact on anything in the world.

The CDC 6600 had a very odd architecture. It had several I/O master processors, but the floating point engine was the slave.

This was prior to silicon microprocessors, so these processors were laid out on circuit boards with arrays of transistors as the active elements. Special care was taken in laying out the wire leads to ensure that signals arrived at the correct time as well as reducing line capacitance.

The wiki describes it as the evolutionary precursor of RISC (like ARM on your phone, for those less familiar).

The CDC 6600 ran at 10MHz.

https://en.m.wikipedia.org/wiki/CDC_6600

The Cray-1 would eventually run the "UNICOS" version of UNIX, and the FPU featured vector registers, which have been brought back in RISC-V.

The Cray-1 ran at 80MHz.

https://en.m.wikipedia.org/wiki/Cray-1

If your supercomputer doesn't come with furniture, is it even supercomputer?

I always like the scene in Sneakers when Ben Kingsley and Robert Redford are sitting on a Cray.

Interesting, cover photo is of a machine that was never sold, only one ever built. (Cray 3)

Just now I was trying to find a microcontroller that would match the performance of the original Cray 1.

But of course, it's complicated.

Throughout of various floating point operations can vary by a factor of 100 on a simple machine. Which is why the standard benchmarks perform a combination of typical calculations. There's the "Gibson Mix" abd "Livermore Loops".

Roy Longbottom is perhaps the most comprehensive compilation of results from testing via these benchmarks.

https://www.roylongbottom.org.uk/Cray%201%20Supercomputer%20...

Essentially, the Raspberry Pi 4 is about 100 times as fast as a Cray 1.

I can find tiny chips that can do specific DSP applications at 160 MFLOPS via fixed-function units, but it's difficult to know if that's comparable.

The FPU on some ARM Cotex-M4+ devices can rival a Cray 1 in general-purpose floating point, and their fixed function units can exceed that 100x. Mostly these are control units for LIDAR devices in automotive applications. They can pull more power than the Apple M1 on my iPad, 20 Watts or so.

How many Cray 1 computers does it take to sit there and listen for me to say "Hey Siri..."?

That last part (The “Hey, Siri! Equivalent) suggests that you may be on the path toward a unit of measurement for want of intelligence.

I think it may be interchangeable with the sqrt(-1).

The Cray PVP line was also doing double precision floating point, and could overlap vector memory operations with math operations. My guess is that you would need a microcontroller operating at several hundred MHz to beat a Cray-1 in practice. The later Cray-1/S and /M variants also supported a 10gbps link to an SSD of several hundred megabytes, which is hard to beat in a microcontroller.

Take a look at the Parallax Propeller 2. Though it probably won't work its ass off on floating point, it is highly truly parallel.

https://www.parallax.com/product/propeller-2-p2x8c4m64p-mult...

If you're into supercomputers like I am, make sure to check the Museum of Technology and Art in Roswell, GA. They have dozens of them, including Crays, on display.

Thank you! I have been fascinated by the life of Cray and his supercomputers. I need to buy this book and read from cover to cover.

I was really hoping this article would have lots of cool pictures.

Reply on news.ycombinator.com