Dreams are coming true – Supercomputing 2016 in Salt Lake City
It was 1951 when Marvin Minsky (1927 – 2016) randomly wired electronic components to mimic neuronal connections. His aim was to build machine capable to learn. Soon the concept was ported to the upcoming digital computer devices, but the so-called neuronal networks did not become useful in practice yet. But they inspired scientific writers and science fiction authors to share their dreams of a world filled with self-learning machines. Artificial Intelligence (AI) was the hype in the 60s and 70s, there were wild speculations about self-driving cars, helping robots and speaking machines. One of the highlights was HAL the selfish computer in Kubrick’s epic movie “A Space Odyssey 2001”.
This years Supercomputing conference (SC 16; November 13th to 18th 2016 in Salt Lake City UT) revealed that some of these dreams have become reachable goals. It was only 4 years ago when Alex Krizhevsky proposed his concept of deep convolutional neural networks, which is used to classified images. A simple personal computer equipped with gamer graphic cards learned efficiently to describe the contents of images. Not only it was able to distinguish between a black cat and a black dog, it was also able to differentiate between different breeds of Swiss mountain dogs. Pardeep Dubey, an Intel Fellow and director of Intel’s Parallel Computing Lab said in his talk at SC16: “I have never seen a technology faster coming from the lab to hit the road.” In her conference keynote Katharine Frase (IBM) focused on the daily use of artificial intelligence in near future notably in health care. On the exhibition floor vendors were dreaming of a gold rush emerging for them from deep learning. NVIDIA is regarding its graphic chips as the leading technology while Chipzilla (Intel) is trying to safe their share of the business by positioning its Xeon Phi many core processors for machine learning. After neuronal networks have reached HPC, machines with hundreds of cluster nodes each equipped with 8, 10 or even more GPUs or other accelerators are fuelling the business. Intel, which acquired Altera in 2015, tries to grasp the lead in another technology, which sounds like a dream: programmable processors. In reality is this technology called filed programmable gate arrays (FPGA) well established, but has never arrived in the field of general purpose computing. A new use case for these chips has been discovered. By wiring a part of a neuronal network as FPGA program and using them to build an HPC system, which is, both optimised for deep learning and energy efficient a new class of accelerated Supercomputers might change the game.
Besides of the disruptive change caused by neuronal networks classical HPC usage is and will be playing a dominant role. Simulations for science and engineering are still the most important use for supercomputers. A long-standing dream is to simplify the writing of parallel code for scientists and engineers. This dream as well has come true, at least for aviation engineers. Vincent Peter (Imperial College London) presented a python framework for solving problems on streaming architectures. This framework generates parallel C-Code, which is able to run on the worlds fastest machines namely on Piz Daint at CSCS.
Which brings us to the important Swiss contributions. 25 years ago Swiss scientists and politicians dreamt that Switzerland should play a leading role in the field of supercomputing. To make this dream to come true, they initiated CSCS, the Swiss National Supercomputing Centre in Ticino. On the semi annual Top-500 list of the fastest supercomputer the upgraded Piz Daint has been ranked as the fastest machine in Europe, for the 7th time in row now. And it turned out to be 2nd on the Green-500 list ranking the most energy efficient supercomputers. Scientists at ETH, EPFL, University of Basel and other Swiss University are playing an important role in developing new application for HPC or optimising widely used code.
There remains the dream to reach more diversity in STEM fields. To challenge the question: “Why are there not more women in HPC?” The diversity track at the conference as well as the Women in HPC events were important steps towards a more divers HPC community. A Swiss chapter of Women in HPC might follow soon – another step to make this dream to come true.