Today Data Vortex introduced its first commercially available HPC system, the DV206, consisting of 32 commodity dual processor servers connected with a Data Vortex Network System and Data Vortex Network interface cards. The capabilities of its new system will be demonstrated at the SC14.
Data Vortex targets “Hard Problems” requiring intense processor-to-processor communications that do not perform well on commodity clusters. While other HPC interconnects load data into large packets to achieve performance, the Data Vortex design embraces small packets. The network system is a high-radix, high bandwidth, low-latency, flat network carrying 64-bit payload packets. It efficiently gathers and scatters fine-grained data, enabling efficient communications among servers. The Data Vortex systems embrace a new computing paradigm to enable a wide class of algorithms. Strong performance on a suite of important problems will be demonstrated on the DV206. The GFFT benchmark will be used to compare Data Vortex systems to existing parallel systems.
The Data Vortex systems represent a step away from the world of LINPACK and MPI. The programming environment consists of C on commodity processors along with a new parallel data passing system. The system comes with a rich API which allows a straight forward intuitive programming environment to realize the strengths of the hardware. This computing environment illustrates a completely new way of thinking about computing and will create new classes of algorithms.