Skip to content

The Computational Power of Camp Nou

June 20, 2014

I joined the Barcelona Supercomputing Center almost a year ago and have always been impressed by the proximity of the BSC facilities to Camp Nou stadium. It is quite an interesting experience working late on a night of the European football Champions League because from my office I can hear the choruses of the supporters and almost see the upper part of the bleachers — such an emotion being so close to one of the largest stadiums in the world.

Camp Nou’s capacity is about 100,000 people. Assuming that the average person in the stadium is carrying a smartphone with them and assuming the performance of the System on Chip (SoC) processor in a smartphone is ~20 GFLOPS, then on the night of the Champions League, you would have almost the same computational power in that stadium as the top entry of the TOP500 list from four years ago. Or, if you’d prefer, it would be the equivalent of about 4 percent of the Tianhe-II supercomputer, the number one entry in the current list.

The computational power that we carry around in our pockets is really impressive, the capabilities and modularity of mobile SoCs are improving tremendously. The mobile market is growing in double digits worldwide and consequently the price per chip is becoming more and more competitive. Are all these factors somehow affecting HPC?

On the one hand, processors within our portable devices are HPC-capable: multi-core mobile CPUs have impressive performance, double precision capabilities and will very soon have 64-bit support. Moreover embedded GPUs offer multiple stream processors and are very often OpenCL capable, just like their bigger sisters in huge data centers.

On a less technical note, the business model proposed by the mobile market is flexible and could affect the HPC world even more. In HPC we are used to single processor providers taking care of almost everything: chip design, test and manufacturing. The successful model that the mobile world is showing is a modular market made of multiple IP providers, producing different IP modules (e.g. CPU, GPU, NIC, etc.), multiple chip integrators and multiple semiconductor foundries.

The adoption of the mobile model would have two main effects, both of which have beneficial consequences. First, competition between producers of IPs, chips and systems speeds up and boosts innovation. Second, modularity provided by the IP blocks allows for easy SoC customization, increasing integration diversification and greater energy efficiency in the final system. This modularity makes de facto custom chips and application-oriented machines more efficient, less exotic, and more appealing than they were in the past.

A sure consequence of all this is that architectural heterogeneity becomes more and more an essential factor — whether this will be a resource to be exploited or a trend to be driven is not clear, but we all will have to deal with this, technically and philosophically. Is the era of monolithic CPU designers/silicon manufacturer ending? Can the mobile ecosystem influence HPC modus operandi? Can the mobile business model overtake the HPC business model?

If I have awakened an interest in these questions, learn more about research on Exascale in Europe and if you want to see the computational power of the Camp Nou full of mobile devices enclosed in a HPC rack please visit the Mont-Blanc showcase within the European Exascale booth #833 or meet us at our joint BoF “Exascale Research: The European Approach” on Tuesday, June 24, 2014, 2.15pm – 3.15pm in Hall 5.

Filippo Mantovani is a postdoctoral research associate of the Heterogeneous Architectures group at the Barcelona Supercomputing Center (BSC). He graduated in Mathematics and holds a PhD in Computer Science from University of Ferrara in Italy. He has been a scientific associate at the DESY laboratory in Zeuthen, Germany, and at the University of Regensburg, Germany. He spent most of his scientific career in computational physics, computer architecture and high performance computing, contributing to the Janus, QPACE and QPACE2 projects. He recently joined BSC’s Mont-Blanc project.