5 ms·

From Megaflops to Total Solutions: Cray and Supercomputing History [pdf]

40 points by poindontcare 10y ago

tcoppi 10y ago
Should be tagged with a date of 1993, it isn't immediately clear what timeframe this was published other than after the "late 1980s".
- daveguy 10y ago
  Most recent reference is 1991 with many from 1990. I think it is safe to say (1991). Edit: includes an internal reference to "late 1991", so maybe (1992) or (1993) is more accurate. Edit2: and you were right (1993). Citation: Boelie Elzen and Donald MacKenzie, "From Megaflops to Total Solutions: The Changing Dynamics of Competitiveness in Supercomputing," in Technological Competitiveness: Contemporary and Historical Perspectives on Electrical, Electronics, and Computer Industries (Piscataway, NJ: IEEE Press, 1993), 119-151.
Sephr 10y ago
Amazing how far we've come. Now you can get 10.6 TFLOPS of FP32 in only 300W (reference: GP100 in the Nvidia DGX-1).
- tcoppi 10y ago
  It isn't a completely fair comparison, but it is indeed amazing that you can buy a GPU for < $1000 that would, going by raw FLOPS, by itself be on the TOP500 in the early 2000s, and we use them to browse facebook and play League of Legends :)
  - TheOtherHobbes 10y ago
    They'll end up in phones and watches eventually, running CNNs for speech recognition and other useful things.
trsohmers 10y ago
If you enjoyed this article, I highly recommend the book "The Supermen: The story of Seymour Cray and the technical wizards behind the supercomputer". It goes into a lot of detail regarding ERA before Cray, the creation of CDC (which is a fantastic start up story regarding how they got funded) along with a lot of specifics on Seymour and his thinking and development style. Seymour is my all time hero, and learning extensively about him and his architectures was what gave me my love of computer architecture and design, with this book being a fantastic information source.
- pklausler 10y ago
  I worked for Cray Research and its successors for many years, including a stint at Cray Computer Corporation in Colorado Springs on the Cray-3 project. I wrote the compiler's optimization and vectorization phase, and this gave me a chance to interact with Seymour on several questions and problems with the architecture. It remains a very fond memory.
simplicio 10y ago
Probably worth noting that the thesis of the article (that network effects and buyer lock-in would keep Cray from scumbing to competitors) didn't really pan. Both Cray founded companies went bankrupt a few years after the article was published.
- aab0 10y ago
  One lesson I take away is how fragile a single-product company is. Each new Cray computer had to be either a total or good success, or the company would die. Like a shark, it had to keep swimming forward, and at Cray-3/4, it lost momentum and asphyxiated. Which is why now IBM builds so many supercomputers...
  - aab0 10y ago
    (Which is not to take away from Cray's accomplishments. It's not every engineer, or every industry, where it's possible to decide to increase the performance of a device by an order of magnitude and do so something like 7 times in a row if you include Cray-3 and Cray-4 which apparently did physically exist, according to Wikipedia.)

40 points by poindontcare 10y ago

Should be tagged with a date of 1993, it isn't immediately clear what timeframe this was published other than after the "late 1980s".

daveguy 10y ago
Most recent reference is 1991 with many from 1990. I think it is safe to say (1991). Edit: includes an internal reference to "late 1991", so maybe (1992) or (1993) is more accurate. Edit2: and you were right (1993). Citation: Boelie Elzen and Donald MacKenzie, "From Megaflops to Total Solutions: The Changing Dynamics of Competitiveness in Supercomputing," in Technological Competitiveness: Contemporary and Historical Perspectives on Electrical, Electronics, and Computer Industries (Piscataway, NJ: IEEE Press, 1993), 119-151.

Sephr 10y ago

Amazing how far we've come. Now you can get 10.6 TFLOPS of FP32 in only 300W (reference: GP100 in the Nvidia DGX-1).

tcoppi 10y ago
It isn't a completely fair comparison, but it is indeed amazing that you can buy a GPU for < $1000 that would, going by raw FLOPS, by itself be on the TOP500 in the early 2000s, and we use them to browse facebook and play League of Legends :)
- TheOtherHobbes 10y ago
  They'll end up in phones and watches eventually, running CNNs for speech recognition and other useful things.

trsohmers 10y ago

If you enjoyed this article, I highly recommend the book "The Supermen: The story of Seymour Cray and the technical wizards behind the supercomputer". It goes into a lot of detail regarding ERA before Cray, the creation of CDC (which is a fantastic start up story regarding how they got funded) along with a lot of specifics on Seymour and his thinking and development style. Seymour is my all time hero, and learning extensively about him and his architectures was what gave me my love of computer architecture and design, with this book being a fantastic information source.

pklausler 10y ago
I worked for Cray Research and its successors for many years, including a stint at Cray Computer Corporation in Colorado Springs on the Cray-3 project. I wrote the compiler's optimization and vectorization phase, and this gave me a chance to interact with Seymour on several questions and problems with the architecture. It remains a very fond memory.

simplicio 10y ago

Probably worth noting that the thesis of the article (that network effects and buyer lock-in would keep Cray from scumbing to competitors) didn't really pan. Both Cray founded companies went bankrupt a few years after the article was published.

aab0 10y ago
One lesson I take away is how fragile a single-product company is. Each new Cray computer had to be either a total or good success, or the company would die. Like a shark, it had to keep swimming forward, and at Cray-3/4, it lost momentum and asphyxiated. Which is why now IBM builds so many supercomputers...
- aab0 10y ago
  (Which is not to take away from Cray's accomplishments. It's not every engineer, or every industry, where it's possible to decide to increase the performance of a device by an order of magnitude and do so something like 7 times in a row if you include Cray-3 and Cray-4 which apparently did physically exist, according to Wikipedia.)