World's Most Powerful Supercomputers

This video is #sponsored by Brilliant.

#MostPowerfulSupercomputers #sideprojects

[RayLabs]

One of my claims to fame is that I was the Product Manager at IBM for Roadrunner from 2002 to 2009 (when I was laid off). Roadrunner was the worlds fastest Super Computer for several years. I was very proud of it, and my entire team that made it happen. It was installed at Los Alamos National Laboratory in New Mexico with final installation in 2008 and decommissioned in 2013.

[jkabel]

As a Product Manager for HPE Cray and directly involved in the building of Frontier, I must say that I LOVE this video! Great Job!

Only criticism I have is the use of the HP Logo and not the HPE Logo in the video. We are two separate companies now having split in 2015.

[slickbillwilly]

It wasn’t a Cray 6600, it was a Control Data Computer (CDC) 6600. Cray was the designer who later left to form his own computer company. (I grew up using the follow on CDC 7600 in high school—CDC was the local computer company so the state education department had one shared by the state’s high schools; and Cray-1 and Cray-2 later in my first post-college job.)

[RichJordan]

Lot more editing time than I assumed. Since it's now October and you mentioned taping in July. Glad the world survived that long.

[QBCPerdition]

Kevin Jennings seems to write virtually half of all Simon Whistler content. Keep up the good work, Kevin.

[DFSJR]

I was a technical engineer at Concurrent Computer Corporation and we made a series parallel processing 32-bit superminicomputers that were mainly used in military flight simulators. These were no were near the power of the computers in this list, but they did there job. I am waiting for the release of "THE SIMON" supercomputer that is running off of Simon's uploaded mega brain.

The Frink prediction was a callback to “I think there is a world market for maybe five computers.” Thomas Watson, president of IBM, 1943

[maxdanielj]

What is also interesting about modern supercomputers is how they're cooled

[StoneInMySandal]

I spent some of the best years of my career at ORNL. It’s a truly amazing place existing in a world almost entirely its own.

[ignitionfrn]

3:00 - Chapter 1 - Top 500
4:25 - Chapter 2 - Summit
6:45 - Chapter 3 - Fugaku
8:40 - Chapter 4 - Frontier
11:50 - Chapter 5 - Honorable mention

[jensenhealeyefi]

Supercomputers are more than just the chips that go into them. The interconnection network is also vital to the performance potential of the system as a whole. The HPE Slingshot network technology used in Frontier is newer generation replacement for the older Infiniband network technology used in previous supercomputer systems.

[alext]

Fugaku's predecessor, K was the last major use of SPARC, Fujitsu's SPARC64VIIfx iirc, and now it uses ARM based A64FX. I remember reading about K in computer magazines as a child (I'm American).

[rashkavar]

Things have calmed down significantly - much as people's attitudes toward Covid have become less panicked in general - but the distributed networks have the potential to get a LOT of processing done when people are suitably motivated to help out.

[jamesdybing]

I was a pipe welder /pipe fitter on the cooling system for the Lumi super computer in Kajaani. It was a very interesting project.

[JeffLarkin]

Nice video. As someone who works in the industry, you did a nice job summarizing a very complex topic.

[ronconte]

20 years. The computing power of a cellphone is equal to the computing power of a supercomputer from ~20 years earlier. From a room-sized computer to a pocket cellphone.

[janedunlap]

Ahh, Cray. My dad used to sell components to them, so I grew up with them as a household name. Then they were bought by Just a shame. Btw, Mr. Cray was a really interesting guy. He dug tunnels for fun.

[rancidbeef]

CDC was never part of Cray Research (or Cray Computer Corporation). It was cofounded by Seymour Cray who later left to found Cray Research.

[godfreja]

Frontier was built by Hewlett Packard Enterprise, which was spun off from HP and is a separate, publicly traded company.

[bensnyder]

To add to the comment about AI compute, AI is normally trained using FP16, a lower precision floating point than the FP64 normally used in supercomputers. Dojo will be capable of 1.1 exaflops at FP16, which sounds impressive, until you realize that Google already has a 9 exaflop TPU system running right now, and Frontier is capable of over 14 exaflops if run in FP16 precision. In terms of comparing to the TOP500, you really can't rank Dojo against those machines, because the Dojo chip doesn't support FP64 compute.