Having spent a week in Dallas at Supercomputing 2018 in November, I returned with a renewed respect for the geeks and nerds that are pushing the envelope with artificial intelligence (AI), machine learning (ML), High Performance Computing (HPC, also synonymous with Supercomputing). I also wanted to understand what is being done with all of this computing power and what is the motivation for all of this private and public institutions to have the biggest boldest supercomputer on Earth?
The team from Pawsey Supercomputing from Perth, Western Australia was kind enough to take me through a whole host of amazing case-studies from protecting the lives of premature babies to making energy generation from waves more efficient. Many universities and international private and public research centers were present, presenting their current capabilities and their plans for the next generation of supercomputers.
One thing was clear very quickly, the next generation of supercomputers will be more powerful (more flops), more efficient (more flops per watt), more expensive and primarily cooled using Liquid Cooling, either direct to chip or liquid Immersion Cooling (Daniele Rispoli and Diarmuid Daltún analised the benefits and challenges of HPC Immersion Cooled in one of our webinars).
What seems to motivate all HPC/Supercomputing teams is to make it (onto and) as high as possible on the Top500 list or the Green500. Although the US holds the top 2 spots on the Top500, China dominates overall with 206 out of the 500 vs. the US’ 124. Most of these installations are funded by public grants in cycles of 3-5 years.
Submer’s CEO Daniel Pope and I had the pleasure of a personal VIP tour of MareNostrum 4, Barcelona and Spain’s most powerful supercomputer, clocking in at 13.7 petaflops (around a tenth of the power of Summit, the current #1 with 143.5 petaflops) but possibly the most beautiful setting, located in a chapel, stained-glass windows included.
The world is now focused on bringing things to the next level with exascale computing. Why is this significant? Well, it is said that this is the level the neural brain operates on. These currently seem on-target for 2020, with China, the US, the European Union, India and Japan in the hunt for that enviable #1 position.
Submer will work with a number of HPC/Supercomputing projects across Europe, Asia and the US in 2019, saving some of them >€10M per year in electricity costs, meaning during their 3-5 year funding cycle, they can save up to €50M total that can be invested in more powerful hardware.