The Next Generation Vector System: SX-ACE
Building on the successful evolution of the SX-series of vector-supercomputers the next generation of this product line clearly focusses on all aspects of efficiency:
- The personal efficiency of the scientist or engineer by providing superior ease of use to achieve the necessary result. This is realized by the provision of a world-record memory-bandwidth per core and the resulting strong single-core performance.
- The high computational efficiency of the machine, which means that it achieves a good sustained performance on real applications, not only on LINPACK. In this regard the new machine drastically exceeds the level of systems built from standard components.
- Ecological efficiency in the sense of superior sustained performance per Watt of power consumption on real applications. This is the major design-target of the machine, and as a result it achieves ten times higher performance per Watt than its predecessor, the SX-9. At the same time the machine features a very compact design, reaching about five times more performance per unit of floor-space than the SX-9.
Consequently the new generation named SX-ACE gives you completely new options for simulations in science, research and development. The fast processor helps you run applications with the highest sustained performance even with a relatively small number of CPUs, while on the other hand the system is designed to scale.
Key Points of SX-ACE
- Performance of 256 GFLOPS per CPU with 64-bit precision. A node is built from one CPU and 64 GByte of memory.
- Four vector cores of 64 GFlops each per CPU, with 256 GByte/s bandwidth to the memory. Each individual core can realize this bandwidth individually.
- Drastically reduced latency of memory access, an important improvement compared to the SX-9.
- The architecture of the core is an enhancement of the SX-9-CPU.
- The second generation of the “Assignable Data Buffer” (ADB, a “vector cache”) with significant enhancements compared to the first implementation in the SX-9. It results in a performance-improvement for real applications.
- A rack contains 64 nodes, 16 TFlops peak, one order of magnitude faster than a rack SX-9.
- Proprietary interconnect technology, so-called IXS, realizing low latency and high bandwidth.
- Hybrid cooling, the CPU is contact-cooled, while the remaining components are air-cooled.
- New parallel and highly scalable filesystem (ScaTeFS) and Batch-Scheduler capable of supporting hybrid configurations made of SX-ACE and LINUX-Clusters (NEC LX-series) to enable the hybrid usage of vector and scalar systems even for most challenging workflows and under critical operational conditions.
Benefits of SX-ACE
- Extremely high efficiency and unrivaled unique sustained application performance
- Fast buffering/Assignable Data Buffer (ADB)
- Hybrid parallelization with multi node configuration
- High Density Packaging
- Optimized for low footprint and low power consumption
- Excellent price-performance ratio
- Easy installation, implementation and use