Redefining Scalable OpenMP and MPI Price-to-Performance with Numascale’s NumaConnect

White Papers > Big Data > Redefining Scalable OpenMP and MPI Price-to-Performance with Numascale’s NumaConnect

Using commodity hardware and the “plug-and-play” NumaConnect interconnect, Numascale delivers true shared memory programming and simpler administration at standard HPC cluster price points. One such system currently offers users over 1,700 cores with a 4.6 TB single memory image.  The NumaConnect cluster excels at both OpenMP and MPI computing within the same shared memory environment. No extra software or program modifications are needed to take advantage of the entire system. Results for the NASA Advanced Supercomputing (NAS) Parallel Benchmarks have set a new record for OpenMP core count and problem size. OpenMP results show good scalability, with best results coming from larger problem sizes.
In addition, NumaConnect shared memory MPI performance delivers better results than InfiniBand clusters, using standard tools without modification for the underlying hardware environment. A cost comparison with a small FDR InfiniBand cluster shows a comparable price point when ease of programming, high performance, and ease of administration are considered. Several production systems are performing satisfactorily, including those in University of Oslo in Norway, Statoil, and Keele University. Finally, we pose the question: If you can get scalable OpenMP and MPI performance, ease of programming, and ease of administration at commodity cluster price points, why limit yourself to an MPI cluster?

Built to work as a “plug-and-play” hardware solution, the NumaConnect from Numascale uses industry-standard HyperTransport to seamlessly join commodity motherboards into a true shared memory system running a single Operating System (OS) image. Numascale’s NumaConnect™ technology enables system vendors to build shared memory HPC clusters with servers at a fraction of the price of enterprise-level shared memory systems. At the heart of NumaConnect is the NumaChip™ that contains both cache coherency logic and switching capability to build large torus topologies. NumaConnect provides a record-breaking main memory capacity of up to 256TB in a single commodity-based system. The following discussion will highlight some of the important advantages of NumaConnect systems and most importantly, provide performance and cost numbers that make NumaConnect-based HPC systems a compelling choice.

Error: Contact form not found.

All information that you supply is protected by our privacy policy. By submitting your information you agree to our Terms of Use.
* All fields required.