The Tera 100 at the French CEA-DAM has become the first European supercomputer to break 1 petaflop.
Europe has a new fastest supercomputer, as the Xeon-powered Tera 100 breaks the petaflop barrier for the first time.
While it's not going to challenge China's Fermi-powered 2.5 petaflop Tianhe-1 supercomputer
any time soon, the joint project between supercomputer specialist Bull and the Military Applications Division of the French Alternative Energies and Atomic Energy Commission is now officially the fastest supercomputer in Europe.
Based around 17,480 Intel Xeon 7500-series processors, the room-filling supercomputer packs 140,000 RAM modules into its chassis in order to offer a total memory capacity of 300TB. For storage, the system is backed by a whopping 20PB - that's petabytes
- of disk capacity, with benchmarks demonstrating data transfer rates of around 500GB/s.
In the Linpack benchmark, which forms the basis of the TOP 500 supercomputer ranking system, the Tera 100 managed to score 1.05 petaflops - the first supercomputer in Europe to do so. Engineers working on the project believe that they can push it harder, too, with an estimated peak processing speed of 1.25 petaflops.
Designed for round-the-clock operation, the Tera 100 is built to offer the widest possible range of simulation applications to its military masters, distinguishing itself from rival supercomputing platforms that are often built around the needs of a single application.
Speaking of the landmark benchmark result, CEA-DAM director Jean Gonnord claimed that the partnership with Bull 'opens up the way for even more powerful systems, and for even greater co-operation in the design and development of the next generation of European computers; the Exa-scale systems that are expected to appear before 2020.'
Are you pleased to see the state of European supercomputing advancing ever forward, or just disappointed to see that it's the military that will be playing with this particular toy? Shocked to see a supercomputing specialist not
going down the CUDA route? Share your thoughts over in the forums