What is your next challenge in HPC

Understanding Supercomputers

Supercomputers represent the next frontier in computing capability and the key to unlocking the future in medicine, clean energy, climate change, genetics, artificial intelligence—all fields of scientific research, for that matter. In recent years, supercomputers have helped mankind do incredible things, such as accelerating the race for understanding the effects of climate change, with the help of systems like Frontier, developed by HPE and housed at Oak Ridge National Laboratory, which are enabling scientists to examine the minute details of climate change, down to the level of how individual cloud droplets contribute to global warming. Others, like Japan’s Fugaku, have been instrumental in disaster prevention, by using its ability to model seismic activity to help predict earthquakes and tsunamis, particularly in earthquake-prone regions like Japan.

Since their inception, the relentless pursuit of supercomputers has come at a price: machines that are not only faster and bigger but also increasingly costly. So much so that accessing and experimenting with these massive machines is akin to winning the golden ticket to Willy Wonka's Chocolate Factory – everyone wants a piece. These High-Performance Computing (HPC) systems are the pillar upon which our current scientists shape the future. But even a bright future has its challenges.

The Moore the Merrier

It’s commonly known that, according to Moore’s law, the number of transistors on an integrated circuit doubles every two years. Interestingly, the number of people declaring “Moore’s Law is Dead” seems to double at the same rate. Only time will tell if Moore's Law will continue to evolve with new technologies and manufacturing approaches, or if it will eventually reach its limits.

Since 1999, processor manufacturers have transitioned from waiting for advancements in new silicon technology to focusing on new architectural innovations. This shift led to the introduction of multithreading, multiprocessing, and increasingly complicated pipeline features – though these innovations sometimes introduced security vulnerabilities– in an effort to squeeze out more instructions per clock cycle.

Now, some 20-odd years later, chipmakers are unveiling new processors that provide unprecedented peak performance. Twice a year, during the Supercomputing (SC) and International Supercomputing (ISC) conferences, the TOP500 organization announces the list of the largest exaflop machines. These machines are often measured in terms of peak performance, a purely theoretical metric. Following almost every lecture during SC and ISC, debates emerge about whether this measurement method drives the hardware vendors to create new tailor-made chips and/or hand-coded software purely to attain the highest score. These are theoretical and philosophical discussions—and some might even argue that they are rhetorical.

But what about real-world computing demands?

There is a good reason for this. While peak performance numbers often grab headlines, they rarely reflect the day-to-day challenges organizations face. In fact, these numbers can be misleading, since they offer limited insight into how systems and applications perform under real-world conditions. HPC customers prefer solutions that can consistently power their diverse workloads with optimal performance and efficiency all of the time. They are growing less and less interested in theoretical benchmarks they rarely will achieve. As a result, HPC customers are demanding a new approach to address their requirements. They seek acceleration architectures designed for their workloads, delivering practical, efficient performance that translates to faster breakthroughs and new innovations.

The HPC community is increasingly aware of these limitations in the field and impatiently seeking solutions. In future posts, we will explore some of the benefits to be gained by rethinking supercomputer architecture from the ground up, and how we can move beyond these restrictions by innovating new approaches that provide everyone the golden ticket to an accessible supercomputer.

About the author:

Elad Raz is the founder and CEO of NextSilicon, a company pioneering a radically new approach to HPC architecture that drives the industry forward by solving its biggest, most fundamental problems.