The race to build the fastest superComputer in the world has been going on for some time.

First MegaFLOP computer - IBM 7030 1961
First GigaFLOP computer - Cray 2 1985
First TeraFLOP computer - ASCI Red 1997
First PetaFLOP computer - Roadrunner 2008
First ExaFLOP computer - Frontier 2022

The list is somewhat linear after the advent of ICs in the mid 70s, with a three order of magnitude increase every dozen ( or so ) years, and we should expect ZetaFLOP performance ( Floating Point Operations per Second ) by the mid 2030s.

The Chinese 'LineShine' System id the current #1, at over 2 ExaFLOPs, with the rest of the rankings in this article

LineShine Debuts at No. 1 as the TOP500 Enters a New Global Exascale Era | TOP500

Maybe @Sensei will be the only one interested in this article, but, as someone who still knows how to use a slide-rule ( after learning in the mid-0s, when a calculator cost $300 ), I am amazed at the progress.

4 hours ago, MigL said:

after learning in the mid-0s,

Sorry, mid-70s .

OK, I've still got my ARISTO-SCHOLAR LL which I bought when starting my studies at Technical College in 1962 - but disuse and old age mean I'd have to brush up on how to use it

I also had a circular pocket slide rule but that has long since disappeared.

I just started to learn to use a slide rule but pocket calculators were becoming a thing in the early 70's.

I was in the last class at the navy's nuclear power school to have to use a slide rule c1977. Of course, the entire fleet had been using calculators for well over a decade by then...

I was in GR 12, in 1975, when my brother ( actually my dad did ), then in Gr 13 ,bought a Texas Instruments SR-51 calculator for $300 Canadian. We 'shared' that scientific ( with red LEDs ) calculator as that was quite a sum when houses cost $25-30000, and a new car about $3000.
And only 5 years after Tom Hanks used a slide rule to calculate the trajectory for the Lunar Lander's return flight after the LOx tank explosion.

No, wait ... that was a movie.

4 hours ago, npts2020 said:

I was in the last class at the navy's nuclear power school to have to use a slide rule c1977. Of course, the entire fleet had been using calculators for well over a decade by then...

I wouldn't imagine a safety-critical discipline, like naval nuclear work, would change procedures quickly.

Edited September 30Sep 30 by StringJunky

21 hours ago, StringJunky said:

I wouldn't imagine a safety-critical discipline, like naval nuclear work, would change procedures quickly.

Ya, but I would think if there was any question about anything like that it would have been done onshore first. The real reason was because calculators were still pretty expensive and students being low man on the totem pole, so to speak, made us a low priority. Was interesting to see some guys (no women in the nuclear program back then) carrying calculators and some with slide rules. It would probably take me a while to figure out how to use it again but it is a pretty ingenious device and definitely takes some knowledge to use.

In its early days, progress consisted primarily of increasing processor frequency. Doubling the frequency meant doubling the speed, and so on.

Until late '80 ALUs and FPUs in either Intel and Motorola CPUs were separate.

https://en.wikipedia.org/wiki/Arithmetic_logic_unit

https://en.wikipedia.org/wiki/Floating-point_unit

So if one wanted to have floating point math, he or she had to emulate it with integer ALUs, which is a very slow process. After '84 in the case of Motorola, they could invest in 68881 FPU coprocessor.

The first Motorola with built-in FPU was 68040 released in 1990 https://en.wikipedia.org/wiki/Motorola_68040

The first Intel with built-in FPU was 486 DX released in 1989 https://en.wikipedia.org/wiki/I486

(In 486 SX it was disabled)

For over two decades, the main driving force has been increasing the number of cores. If you double the number of cores, you get double the speed, etc. But this requires that your algorithm scales well in a multi-thread environment. Which isn't always possible. Sometimes it's never possible, because, for example, the result of the previous calculation is a parameter to the next function.

If the goal is to calculate something as quickly as possible, instead of writing a software program for a standard CPU, you can implement the algorithm directly into the hardware logic using an FPGA. You're essentially creating your own private processor instead of making software.

https://en.wikipedia.org/wiki/Field-programmable_gate_array

The cheapest models on AliExpress cost $30-40. Search for the Tang Nano 4K/9K for example.

Is it even a computer if you have, for example, a cluster of machines doing one task? After all, you can ask people (billions of them) to download some app on their phones/PCs and send them a task to do when those phones/PCs are in idle mode, and can you call this a "supercomputer" - a mesh, grid, of machines connected via the Internet?

Hackers, for example, mine Bitcoin. Someone visits their website, the browser starts calculating their code, multiplied by thousands or millions of visitors..

On 9/27/2026 at 11:18 PM, MigL said:

First TeraFLOP computer - ASCI Red 1997

My processor in this laptop has 0.5 TFLOPS. I'll turn on the second laptop and we'll have 1 TFLOPS.. ;)

Edited October 19Oct 19 by Sensei