Arguably, a contemporary smartphone will typically have more processing power than the Space Shuttle originally did, though the later upgrades close the gap quite a bit. When first launched, the shuttle five guidance computers contained close to half a megabyte of core memory each (those had to be enormous storage racks). Add in the GRiD laptop, you are getting kind of close (bubble memory? holy crap). Still, what the shuttle’s computers did was pretty highly specialized and relied heavily on redundancy: if your cell phone were running shuttle guidance and experienced a failure (crash), it might have trouble recovering in time to save your ass. Though, I suppose 5 phones working together would do the trick and still be less expensive.
Here’s a cycle-accurate recreation of a Cray-1A on a single FPGA chip:
http://www.chrisfenton.com/homebrew-cray-1a/
Not quite the same thing as comparing it to a Dell, but it does show that modern computer hardware can fairly easily recreate the function of a Cray 1A.
I’ve had dual 1.5-terabyte drives in my computer for almost 3 years now. Now you can buy 4 TB in a single 3.5" enclosure.
I’m curious about the supers that aren’t on the top500 list, the ones at NSA or other agencies.
Has anyone ever heard of any kind of performance from these? I assume they have ones that are more powerful than the top of the top500.
There are great answers already in this thread but I think the following answer about why some video cards are better at Bitcoin mining than a CPU explains well how comparing different computer architectures can be apples and oranges (video cards in modern PCs work more akin to a super computer than a CPU in a modern PC does):
No, they don’t. The NSA etc. haven’t made their own supercomputers since the 1950’s. They discovered that the machines they designed & built cost much more than estimated, and ran slower. So in the early 1960’s they switched to buying the newest commercial machnes. Like the CDC 6600, 7600, and then the Cray machines. So by the time the Top 500 list was started, NSA was just using commercial machines (but often the 1st buyer of a new machine, and the most powerful version available). These machines appear in the upper range of the Top 500 list, with sites like Lawrence Livermore, Los Alamos, Oak Ridge, etc. – it’s likely that NSA had a similar machine operating, and might have even had it earlier8. But such machines weren’t public to be listed in the Top 500.
- It’s commonly reported that there was a battle between Lawrence Livermore Labs and Los Alamos Labs to get the Cray-1 serial number 001 machine (won by LLNL). But a recently declassified report on NSA computers claims they had ‘the first’ Cray-1 produced. Was there a serial number 000 Cray machine?
NSA had a close relationship with Seymour Cray. His machines had a very RISCy design (before the term RISC was even recognized). But while he was eliminating many instructions from his machines, he added one (PopCount) that seemed way too infrequently used to deserve a place in the machine’s instruction list. But that one just happened to be very important to NSA and cryptology work. So it was there.
An old college of mine (who did work for defence agencies) once observed that there was a “black” Top-500 equivalent list, and it looked mostly like the open Top-500. Another guy I knew was a hardware maintenance guy for Cray, and lived in a city where there were no known Cray machines installed. But there were lots of government departments. It isn’t just spooks decrypting communications. Classified work is going to occur in just about every scientific and engineering filed as well, and those projects will require access to similarly classified facilities - if only to maintain the secrecy of their own existence.
Thanks t-bonham@scc.net and Francis Vaughan, good info.