General statistics
List of Youtube channels
Youtube commenter search
Distinguished comments
About
Lepi Doptera
Asianometry
comments
Comments by "Lepi Doptera" (@lepidoptera9337) on "The Rise and Fall of the Cray Supercomputer" video.
No, not really. It was Moore's law that put an end to custom machines. I was working on a high energy physics project in the early 1990s that used field programmable gate arrays for performance reasons (we had GByte/s data coming in at packet rates of 60MHz). By the time we were done it would have been easier and cheaper to implement the entire crate of electronics on standard CPU boards. It also would have made it a lot easier to program. Today I could write the entire algorithm stack in C or C++ in a couple weeks and run it at 10x required speed on the GPU of my laptop. Back then it took a dozen physicists and EEs to build the hardware and write the VHDL code.
2
I can buy the same amount of computing power from Google and Amazon for a couple thousand dollars per month that would cost me millions if I had to buy IBM hardware. The difference is that I don't have to pay for what I don't use. The only reason why banks etc.. are using IBM mainframes are reliability and security. People who aren't running that kind of business process don't need it.
1
The users of supercomputers typically still do. Apart from the numerical libraries that are being shared and a few fairly basic OS components for scheduling all the code on a supercomputer is pretty much tailored to the user's needs. That's also a lot more interesting that full stack design, where every programmer basically reinvents the same old boring user interface, database and backend code over and over, again.
1