2018 PREDICTIONS: THE EXAFLOP BARRIER HAS ALREADY BEEN BROKEN
When starting this article, I have felt a bit of an “Aramis Fuster” complex, giving predictions for next year, at this time, that when we want to remember we will be eating nougat. But this prediction is real, made from the field of computing and more specifically from what has been called HPC, High Performance Computing, High Performance Computing or Scientific Calculation, of which our company Sistemas Informáticos Europeos, is a specialist.
When the consulting firm Gartner Group predicted that by 2020 our researchers would be requiring exaflops for their calculations, we all had a hard time believing this prediction. Especially since it meant going from a few Pflops, offered by the computing centers, to multiplying this figure by 1.000. We are talking, neither more nor less, than a trillion trillion calculations per second.
Only in 2008, Petascale computing began to work, that is, in the order of Pflops. This new frontier is from Exascale.
However, the predictions fell short. For the first quarter of next year, SIE will launch, together with Gigabyte, a platform that will allow us to obtain up to 0,5 Exaflops in a single machine and with a Mellanox Infiniband or Intel Omnipath connection at 100 Gb/s, we will be able to reach the Exaflops with only two teams.
In fact, SIE launches this month equipment based on the PCI version of the new NVDIA Volta V100 cards, which are already available and offer up to 120 Tflops.
Computer science surprises us once again and the exponential progress that companies such as NVDIA, Mellanox, Intel, Gigabyte, etc. are making in recent years, leave us speechless.
It seems incredible, but it's true. We had already been told that in less than 10 years, the world would change more than in the last century and it seems that it will be true. If Alan Turing raised his head, he would die of fright again. Go from here, our appreciation to the father of modern computing, famous for having cracked the secret Nazi codes contained in the Enigma machine and who died ostracized for his homosexuality in 1952, after saving an estimated 14 million lives in Europe. . If you haven't seen it, be sure to watch the movie The Imitation Game (Cracking Enigma), with a stellar performance by English actor Benedict Cumberbatch.
But let's get back to the topic at hand. This new computing power opens up almost infinite possibilities in the field of Deep Learning and Machine Learning and allows dreaming at a research level in issues such as cancer, autonomous driving, medical imaging, pattern recognition in images, etc. we could never have dreamed.
Gentlemen, buckle up, because by 2030, there are already talks of hundreds of Exaflops and advances that we could not have imagined just 10 years ago.
However, all this that seems so modern dates from the 40s. At that time, the theory of neural networks and their algorithms had already been proposed, but there were no teams that could execute it effectively and they had to wait more than half a century to see the light.
Finally, 2018 (two years in advance) will be the year of Deep Learning, Machine Learning and Big Data, as we have never known it.
To give an example, the Gaia space probe, from the European Space Agency, which was launched in 2013, where Spanish Universities are involved, will send us millions of Gigabytes of information and will obtain a catalog of approximately one billion stars, which of course later it will be necessary to analyze with Big Data solutions.
Another example, hyperspectral cameras (which are capable of photographing from the infrared to the ultraviolet spectrum), instead of the classic RGB image (with 3 positions), generate a thousand positions per point, with its corresponding value scale. Thanks to the real-time analysis in HPC, we will be able to visualize during a cancer operation, which areas have damaged tissue or obtain in real time recognition of objects photographed from the air, which is an anomaly.
Finally, before 2020 we will have the new 5G network. Multinational operators such as Telefónica or
Vodafone, they are already working on it. The phenomenon called IoT (Internet of Things or Internet of Things), will allow devices to talk to each other.
There are currently 7000 billion devices connected to the cloud globally. It is expected that with the rise of the IoT there will be 20000 billion connected devices in 2020 and the forecast for 2025 is 100000 billion small devices connected to the Internet.
For this new phenomenon, real-time Big Data analysis will be necessary. To give a practical case, autonomous driving cars will not only carry their own sensors, but will also communicate in real time and autonomously, with traffic lights, traffic signals, light panels, weather forecast services, parking, speed beacons. , etc (generating tens of Gigabytes of information per hour) and we will see a network with a latency of less than 10 ms and speeds of hundreds of Mbytes per second.
Systems in the cloud will receive all this data and analyze it in real time, to allow road accidents to be a thing of the past (in Spain, every year, road accidents kill more people than cardiovascular diseases).
An era opens that must be recognized is a bit dizzying, but from SIE we find it hectic and that fills us with joy to be able to live from the world of HPC
Raul diaz
Project Manager
Sistemas Informáticos Europeos