Supercomputer promises acceleration, optimisation & flexibility

The Hi√e HPC systems are optimised for accelerated workloads, offering excellent performance and energy efficiency. This is obtained by a high degree of modularity in conjunction with the water cooling of all components. The system is configured to adapt to different applications, avoiding unnecessary components and focussing on the choice of elements, providing maximum performance to the specific customer workload.

One system module can host a combination of ARM-64 or x86 CPUs, Intel or NVIDIA GPU accelerators, as well as other PCIe components. Modules are innovative form factor, hot water cooled enclosures supporting different configurations of the components. They provide computation and control functionality (Intel or ARM processors), acceleration (up to five Intel Phi or NVIDIA GPU accelerators), Infiniband network and additional optional functionalities of storage and visualisation. These modules are logical nodes of a large system, hosted in the Aurora Hi√e rack in any combination up to 128 per cabinet. The High Speed Interconnect allows scaling to any size system with density up to 750TFlop/sDP/m².

The combination of low power processors (Intel Haswell E3 up to 80W TDP or Applied Micro X Gene ARM 64b) and 'energy aware' design makes the system extremely energy efficient, with theoretical values of 5GFlops/W. In addition, the Hi√e is entirely hot water cooled, allowing the highest energy efficiency not only at machine level, but also at datacentre level. In this way, a datacentre could aim at a PUE (Power Usage Effectiveness) around the 1.05 mark.

The Aurora Hi√e systems are equipped with the 2nd gen of the acknowledged Aurora Direct Hot Water Cooling, a technology that puts the coolers in direct contact with components like processors and memories. Compared to the previous cooling generation, the technology has the benefits of lighter and more compact cooling elements, allowing extremely dense packaging inside a computational module of only 130x105x325mm.

To take advantage of GPU/Phi acceleration on Intel and ARM architectures, Eurotech is working to validate and tune the Hi√e configurations for a wide range of accelerated applications in a number of market segments, including high energy physics, bioinformatics, molecular dynamics, CAE, machine learning, finance, GPU-based rendering and seismic migration.

The Hi√e configuration with Intel E3 Haswell 3rd-gen processors and Intel Phi co processors is available for ordering, while additional configurations with Intel or APM X-Gene ARM-64 and NVIDIA Tesla K40 GPU accelerators will be available for ordering in early-to-mid-2015. The development kits of all of the mentioned configurations will be available in early 2015.

“Eurotech’s Aurora Tigon systems are well known for their industry-leading energy efficiency” says Fabio Gallo, Eurotech HPC BU managing director. “With the introduction of Aurora Hi√e Eurotech is yet again raising the bar in accelerated computing, delivering a whole new level of energy efficiency and density. Aurora Hi√e systems are a bold step in the path leading to technologically viable and affordable Exascale systems”.

To satisfy other markets, Eurotech engineering services combine a variety of out-of-box modules to provide systems that fit specific application needs.