HiVe: the green supercomputers to win supercomputing challenges

In a previous post, I highlighted the importance of High Performance Computing (HPC) as enabler of  human progress. Some of the most relevant scientific discoveries and technological advances rely on HPC. Overall, we can state without exaggerating that HPC has an impact on our everyday life.

This is all good and glittering, but it doesn’t come for free.

Science and technology have become progressively more computation hungry, finding in simulation the way to solve bigger problems better. At the same time, the incredible proliferation of data has required HPC technology be used in many data intensive applications like big data analytics (yes, I used the “big data” buzzword…).

This has led nations and organization to procure and set up huge supercomputing installations that have a considerable impact in terms of energy consumed, space occupied and generated waste. In other words, HPC carbon footprint, as for the rest of IT, is becoming a major concern. This is the not so bright side of an otherwise amazing story.

Can we accept that an industry which is  enabling a better future for humankind can actually contribute to put this same future at risk? The answer is most probably no. It is not a case that the HPC community has set the goal to build a one Exaflops system ( a system that performs at one billion billions floating point operations per second – a number good for science fiction novels) within the power budget of 20 mW. This looks like a huge number, but considering that the projection based on current systems is of 500 mW for an Exaflops installation, you may understand the size of the challenge: to give a rough idea, 500 mW is the average power consumption of a 700.000 people city!

Energy consumption isn’t the only challenge. Aspects like energy mix, space usage and waste also contribute to carbon footprint and they are intimately linked to HPC. With the current state of the art, an Exaflops system is set to occupy almost 15000 m2, this is like 2 football pitches only to house the actual machines. Double this space to accommodate  the entire data centre and you get an idea of the amount of concrete and other materials needed.

Eurotech is deeply involved in exascale research. We contribute to it leveraging our design and engineering capabilities to make supercomputing systems as energy efficient and compact as possible.

Take for instance the Eurotech Aurora HiVe: it is a completely new concept in HPC. It was designed with the idea that an essential fit for purpose bare bone computer would perform extraordinary well and efficiently processing specific workloads. At the same time, we did not want to pigeon hole it to only a limited set of applications so we designed it with modularity in mind. The system can be configured with different components (like GPU, accelerators or any other PCIe card) to form modules that can accelerated applications in the most energy efficiency and cost effective way, choosing the most optimized configuration in order to avoid paying for IT, energy and space you are not fully utilizing.

We could reach some remarkable results. For instance the system can deliver more than 1 Petaflops double precision in just 1m2. This means the possibility to theoretically build an Exaflops system within 1500 m2, a much more manageable space.

The HiVe can perform at more than 6 GFlops per watt (sustained). This puts HiVe in the top 10 of the most energy efficient systems in the world according to the Green500 (https://www.green500.org/ ). it means that the same Exaflops system would consume a peak of 166 mW of power, still a lot higher than the 20 mW target, but certainly going into the right direction.

Recently we announced the availability of our HiVe Development Kit, a desktop like system that makes it possible to easily try the Aurora HiVe technology: goo.gl/d0vj9J

#Hi√eAccelerates