Seymour Cray and the Development of Supercomputers | Hacker News Reader

Seymour Cray and the Development of Supercomputers

linuxvoice.com

62 pointsbenev18 comments

I remember the days when it was a point of national pride to own a Cray supercomputer. In the 1980s, India tried to get access to a Cray XMP-24, but was denied access due to dual use capabilities. Instead, they were only allowed to purchase the lesser capable XMP-14. The purchase came with the stipulation of having to host US personnel onsite to monitor the usage of the supercomputer. This denial/control of access also led to the birth of India's own supercomputer efforts by CDAC. Many developments in India's space, nuclear and computing efforts mainly happened due to denial of technology such as Cryogenic engines, supercomputers etc.

Recommended reading for all Cray fans http://www.amazon.com/Supermen-Seymour-Technical-Wizards-Sup...

Came here to recommend this book. It's very good.

A wonderful book. It's light on technical detail—well, void of it—but covers the human story and business background well, and the material is so fascinating, it's hard to believe how fascinating it is. Especially the origin story, which is like the early history of Silicon Valley in an alternate universe of the Midwest. People bringing their dogs to work. CDC getting started by selling stock out of a station wagon. Cray moving his lab further and further away from the managers. It's all so archetypal, most of all Cray himself.

A pattern that recurs through the book is that Cray's achievements worked out best when he collaborated with the much lesser-known Les Davis, who was a master of organization and execution. A good local newspaper article from Davis as of a few years ago: https://news.ycombinator.com/item?id=10334470. I hope someone has gotten a full oral history from Mr. Davis.

Growing up in the 80s, the Cray 2 was my poster computer. Until the day came when I realized it was only 800MHz and had 4GB RAM, and my phone was faster than that.

Hold on there cowboy, the Cray 2 did 1.9 GFLops [1] but todays phones peak at 0.3 GFlops[2].

[1] https://en.m.wikipedia.org/wiki/Cray-2 [2] http://www.greenecomputing.com/apps/linpack/linpack-top-10/

“Anyone can build a fast CPU. The trick is to build a fast system.” - Seymour Cray

The iPhone 6S is rated at 172.8 GFlops. https://en.wikipedia.org/wiki/Apple_mobile_application_proce...

I often thought this, but a Cray is not a normal computer. From some comment, I learned that it had a 8MB L1 cache, meaning it could crunch all of it in one go, then refill, so on and so forth. It means these little 800MHz were used to their fullest, while nowadays multicore GHz chip have to wait a lot to get new data in smaller quantities. Making a Cray still as fast today (for adequate tasks of course).

So todays processor have higher peak bandwidth, on average, Cray can sustain larger bandwidth.

ps: https://archive.is/FWzLF read jojomonkeyboy comment

Most of the Gflops in a phone are in the GPU - precisely because it's massively parallel, like the Cray.

Intel's i7s and Xeons have a more general purpose architecture with much less parallelism, but still manage a steady few hundred Gflops on Linpack.

True, I was speaking about the CPUs. Also GPU are still dedicated to graphical workload on smartphone, or are they used through an opencl-like lib ?

Depends on the GPU. PowerVRs are remarkably general for instance.

There are programming techniques where you offload all the computation onto (via OpenGL!) onto the CPU. I did this for image processing where we needed to do histogram normalization. It was not fun.

But the GFlop numbers given are maybe an order or two of magnitude off from achieved performance for something like Lapack.

I never thought about it that way (I find supercomputer comparch design fascinating for the tradeoffs chosen more than anything else), but the RISC/CISC alternatives can be thought of as an optimization for main memory latency.

If memory access is fast compared to the CPU, RISC designs are more optimal (as you can increase the CPU frequency). If memory accesses take longer than CPU execution, then CISC designs start to make more sense (do more complex things in one go once the CPU is done waiting on things to finally arrive).

One thing that strikes me about Cray's designs is that they look so extraordinarily cool, but the look of the machines is - according to Wikipedia - often the result of technical/engineering requirements/decisions. They were led by technical priorities almost exclusively, yet they came out not only the fastest computers on earth (at the time), they also looked ... awesome.

I always found it somewhat humorous that Cray invested so much effort into the look of it's supercomputers, considering very few people would actually see the inside of a super computing center.

Cray 1: 80 MFLOPS and 8 megabytes in 1976. That was the standard for "mini-supercomputers" in the 1980s.

And subject to ITAR export regulation, as you might imagine.

By the mid 80s, personal computers were starting to get close to these limits and the US government had to start tweaking the rule's thresholds.

Reply on news.ycombinator.com