Supercomputers are getting greener. That’s because with exascale systems on the horizon, representing a 1,000-fold performance improvement over today’s petascale computers, energy efficiency has turned into a priority for the high-performance computing (HPC) industry.
Barcelona Supercomputing Center (BSC) is doing pretty well in the green stakes. Its MareNostrum 4 Power9 cluster ranks ninth in the Green500, which lists the top 500 supercomputers in the world ranked on energy efficiency.
Its energy efficiency has nothing to do with its location in a 19th century church. Although the building is naturally cool, the computer of course needs extra refrigeration to prevent its microprocessor cores from overheating.
The Power9 cluster, which was launched at the end of May 2018 and is mainly aimed at artificial intelligence, has shown itself capable of executing 11.865 x 109 floating-point operations per watt of consumed energy, or 11.865 gigaflops/W.
With that performance, the MareNostrum 4 cluster, based on IBM Power9 processors and Nvidia Volta GPUs, emerges as the greenest machine in Europe.
SEE: Cloud v. data center decision (ZDNet special report) | Download the report as a PDF (TechRepublic)
The most energy-efficient supercomputer in the world is Japanese Shoubu system B, achieving 18.4 gigaflops/W.
The use of the performance-per-watt metric, which measures the rate of computation that can be delivered by a computer for every watt of power consumed, may be controversial as it’s linked to the Linpack benchmark, used for the TOP500 list.
This benchmark doesn’t reflect the overall performance of a given system but a peak performance. Does this mean that running a different program could deliver different results? Probably, acknowledges BSC.
Unsurprisingly, the rankings in the TOP500 and Green500 lists do not match up, which suggests much work still needs to be done before performance goes hand in hand with sustainability. Systems architecture appears to be a feasible way to change this situation.
The top three positions in Green500 are taken by Japanese supercomputers based on the ZettaScaler-2.2 architecture using PEZY-SC2 accelerators. Yet, these machines appear very low in the TOP500 list.
By contrast, MareNostrum P9 CTE’s architecture is similar to the one used by US Summit, ranking fifth in Green500 list and first in TOP500 list. They both employ NVIDIA GPUs. On the other hand, Summit requires 8,806kW while MareNostrum Power9 uses 86kW.
Besides Power9, MareNostrum is formed of two more clusters of emerging technologies: a cluster made up of Intel Knights Hill processors and another one formed of 64bit ArmV8 processors in a prototype machine, using state-of-the-art technologies from the Japanese Post-K supercomputer.
SEE: Special report: The future of Everything as a Service (free PDF)
In April, HPE announced the deployment of the largest Arm-based HPC installations in the world, and described it as a low-energy consumption system. Now that the current challenge is to build exascale systems that consume only 20 megawatts of power, the plans are worth monitoring.
Meanwhile, big internet companies are going their own way in testing new ways to cool down servers through machine learning and investing in green energy sources.
Isaac Hernández, county manager of Google Cloud in Spain and Portugal, points out that Google is committed to become a zero carbon emissions company.
“A Google’s data center uses 50 percent less energy than a conventional data center. So, the utilization that an average user makes of all Google services in a month is equivalent to the consumption that a car makes when travelling 1.5km [about a mile],” he tells ZDNet.
In addition, the internet giant describes itself as the company that buys most green energy in the world. “We currently have contracts to acquire three gigawatts (3GW) from renewable energy projects,” adds Hernández.
A July report from Bloomberg NEF suggests that global investment in clean energy for the first six months of 2018 was worth $138.2bn.
Previous and related coverage
HPE announces world’s largest ARM-based supercomputer
Astra will deliver over 2.3 peak petaflops of performance, which should put it well within the top 100 supercomputers ever built.
How supercomputing power is helping with anti-pollution plans like city-wide car bans
A modeling tool developed at Barcelona’s Supercomputing Center is busy predicting levels of atmospheric pollutants in Spain, Europe, and now Mexico.
Supercomputers: All Linux, all the time
The latest TOP500 Supercomputer list is out. What’s not surprising is that Linux runs on every last one of the world’s fastest supercomputers. What is surprising is that GPUs, not CPUs, now power most of supercomputers’ speed.
HPE to build supercomputer for federal renewable energy research
The supercomputer, dubbed “Eagle,” was designed for the US government’s only lab dedicated completely to energy efficiency and renewable energy.
MareNostrum 4 supercomputer now runs at 11 petaflops but there’s more to come
The ambitions of Barcelona’s Supercomputing Center stretch well beyond the newly-operational MareNostrum 4.
Using ARM chips and Linux, Barcelona center dreams of being ‘Airbus of supercomputing’
A chapel in the heart of Barcelona Univesity is home to one of Europe’s most powerful supercomputers – and a mobile chip-based successor is under development.
Raspberry Pi supercomputers: From DIY clusters to 750-board monsters TechRepublic
The Pi clusters that push the $35 board to its limits.
Arm chips, like those in your phone, will power massive Astra supercomputer CNET
Hewlett Packard Enterprise’s Astra machine will study nuclear weapons’ safety and reliability. It’ll be fast, but not the fastest.