Podcast: A Look inside the El Capitan Supercomputer coming to LLNL

Print Friendly, PDF & Email


 
In this podcast, the Radio Free HPC team looks at some of the more interesting configuration aspects of the pending El Capitan exascale supercomputer coming to LLNL in 2023.

Dan talks about the briefing he received on the new Lawrence Livermore El Capitan system to be built by HPE/Cray. This new $600 million system will be fueled by the AMD Genoa processor coupled with AMD’s Instinct GPUs. Performance should come in at TWO 64-bit exaflops peak, which is very, very sporty. The new box (more like a room) will be 10x faster than today’s fastest supercomputer and faster than the top 200 supercomputers in the world – combined. As the show continues, we talk about the specifics of the system and components. Henry make the unfortunate mistake of bringing up IEEE floating point and sending Dan into a mini-rant. Back to the show, the system should require somewhere close to 30MW worth of electricity, which is much lower than the nearly 60MW predicted just a year or so ago. Not surprisingly, the system will be liquid cooled, but not, as we speculate, cooled by Slushy machines. We have a tremendous tech talk around the varying aspects of the machine and AMD’s great progress in clawing their way back into the market.

Other highlights:

Sign up for the insideHPC Newsletter