Numascale today announced that it will officially demonstrate a live SMP cluster based on AMD Opteron processors and NumaConnect SDK at the International Supercomputer Conference (ISC10) in Hamburg starting May 31st 2010.
The breakthrough technology developed over the last 4 years is being showcased live for the first time. In booth 734 at ISC, Numascale features a multi-node system with each node containing two 6 core Opteron processors. With the FPGA based NumaConnect SDK installed in each node, it extends the cache coherent Hypertransport protocol to the whole system enabling it to run one single Linux operating system image with all memory, all processors and all I/O shared and available.
This is a proof-of-concept built to demonstrate the unique capabilities of the chip based NumaConnect SMP which is due to hit the general market this fall. NumaConnect SMP is a combined high performance interconnect and a cache coherent NUMA engine which enables the building of true symmetrical multiprocessing systems (SMP) using standard off the shelf x86 servers. NumaConnect SMP includes full support for virtualization of processing, memory and IO resources and can be used with standard operating systems like Linux, Windows and Solaris. It is designed for high scalability with up to 4096 servers with 256TB of shared memory in one single system. With up to 16GB of cache memory on each NumaConnect SMP card, it can serve as a platform for the most compute- and memory intensive applications in academia, research and industry.
Currently Numascale has its NumaConnect SDK under implementation with two pilot customers, with several other international customers lined up.
Statoil is one of the world’s largest oil exploration companies, and has applications with extremely large memory requirements, specifically within reservoir analysis and a corresponding interest in the capability to directly address and access very large amounts of data. With the NumaConnect technology, applications can have effective and cached access of up to 256 TB of main memory data.
The University of Oslo plans to build a new large supercomputer for scientific applications with a strong focus on efficiency and total cost of ownership. Efficiency is particularly important since large compute clusters have high requirements for power and cooling. Numascale’s technology will be tested and qualified for large-scale operation in a smaller scale prototype.
“It is a very exciting time to bring this breakthrough technology to market,” says Kåre Løchsen, CEO of Numascale. “For years the HPC market has been forced to pay a very high premium for large scale shared memory SMP systems. With NumaConnect we will enable building of systems that have similar specs as competing systems, but with a price of 1/20th to 1/30th of its enterprise alternatives.”
Target market for NumaConnect is initially the traditional High Performance Computing markets in academia and research laboratories globally in addition to industrial customers with memory intensive applications within Oil & Gas, Energy, Health and Automotive.
“Our sales model should be very compelling both to end users, integrators and OEM’s as we intend to ship NumaConnect at a component level for utilization within a large range of systems and application areas,” continues Mr. Løchsen. “By not shipping systems, our plan is to engage the whole HPC community and allow our end users to work with their favorite vendor or integrator to solve their most complex problems with the lowest possible TCO. “