Building El Capitan: How LLNL’s Exascale Supercomputer Came to Be

When Lawrence Livermore National Laboratory’s next-generation supercomputer El Capitan is deployed in 2024, it will likely be the most powerful computing machine in the world, delivering more than two quintillion floating-point operations per second (2 exaFLOPs) in service of its national security missions.

Once fully operational, the National Nuclear Security Administration’s first exascale supercomputer will perform critical modeling and simulation functions to support America’s nuclear stockpile. But a system as large and complex as “El Cap” doesn’t magically appear overnight. It takes years of planning and preparation, and hundreds of employees at LLNL and the Lab’s industry partners, to lay the groundwork for a machine comprising thousands of compute nodes and requiring as much energy as a medium-sized city.

Over several years, teams have prepared the infrastructure for El Capitan, designing and building the computing facility’s upgrades for power and cooling, installing storage and compute components and connecting everything together. Once all the pieces are in place, the life of El Cap as world-class supercomputer begins. Through the lifespan of the machine, Lab employees will need to operate, troubleshoot and maintain El Capitan around the clock, seven days a week, ensuring that scientists, physicists and code teams can perform their calculations efficiently and in a timely manner.

LLNL-VIDEO-855801

[camofelix]

it's COMING!

Can't wait for details

[Reubenintexas]

I got a tour of Stampede a number of years ago in Austin Texas; el Cap would be nice to see once really buzzing!

[davidgrisez]

In this video they mention bringing 85 megawatts of power into building 453. I worked for 34 years in Los Angeles DWP generating stations, I operated units that at full power could generate 185 megawatts of power. So 85 megawatts of power is a tremendously huge amount of electrical power. I suspect that one challenge of these supercomputers will be building supercomputers that are more efficient in their electrical power consumption. Also it is not just supercomputer drawing tremendous electrical power it is also the tremendous amount of air conditioning and water chiller electrical power to keep everything cool.

[Hawk]

This feels super surface-level. I want more! It gives me factorio vibes, too.

[Sagittarius-A-Star]

I can't imagine how the jobs are split onto thousands of CPUs/GPUs - this must be a science by itself.

[willhaslem]

This will be in your school library in 20 years.

[wileamyp]

When can we get early access to El Cap? It's tough to design software for MI300A when you only have MI250X around...

[iOnRX]

i assume el capitan is working on a super soldier serum?

[Burningwithecstasy]

got any extra PC parts laying around? My gaming machine died, last month

[BasVossen]

what OS does it run with? Any info on CPUs/GPUs?

[sundareshvenugopal]

It would still not be enough to run any version of Microsoft Windows ® or any other
look-alike Operating System.

[laurelcreek]

Where do old supercomputers go to die?

[DaS-jpwn]

2 Exaflops and in a few years
when FU Berlin and the MIT is right in a few years even Laptops with the new ONN-Chips were they have build allready the prototyp, will reach 160 Petaflops (80 terraflops best graphic card x 2500 - the ONN-Chip prototyp allready was 20-25 times (FU Berlin says 20 Times, MIT says 25 times) faster then the fastest normal chip of the same size and 100 times less energy consumation, and the university FU Berlins says its easy to speed this up, up to 100 times faster with a higher laser frequency. means 25 x 100 = 2500). Means it would need in a few years only 20 ONN Chip Laptops to reach 3, 2 Exaflops. It will be a new milestone. When 2 or 3 workstations in a few years reach the computing power of he El Capitain which costs hundreds Millions Dollars. The beginn of a new Era. And they said its easy to build, they used only thinks that allready exist so in a few years the massproduction can start.

[alanmarston]

And I am working with a 6 year "OLD" animal.

[Kenneth_James]

This video gives us nothing. Get a little technical

[rupert]

It is funny that it already seems dated.

[rherman]

Not once did they provide the cost of worse, they kept this vague, so as to not actually tell us anything.