HP, Intel Partner to Expand HPC Into New Areas
In addition, Intel expects more than 100 switch and server platforms to embrace the new Omni-Path Architecture 100 Series, with deployments coming starting in the fourth quarter. Wuischpard said Intel will have a presence at the ISC show beyond its alliance with HP, using the event to give more detail about its HPC efforts and its HPC Scalable System Framework. One of Intel's goals for the initiative is to move the development of HPC-level systems away from the traditional one-offs for particular customers and toward a platform of technologies that can be used by for a broad range of applications. Intel initially introduced the Xeon Phi many-core chips as co-processors that could be used in similar fashions as GPU accelerators from Nvidia and Advanced Micro Devices—as a way of boosting the performance of HPC systems while keeping down power consumption. With Knights Landing, the company is enabling the Xeon Phis to be used as primary processors. The chips will have up to 72 cores, with two vector processing units per core, with up to 16GB of integrated memory and five times the bandwidth of DDR and two integrated Omni-Path fabric ports. Intel will demonstrate the Knights Landing chips during the ISC.Intel also is adding a range of new features to the Cloud and Enterprise editions of its Lustre software, including support for over-the-wire and storage encryption and support for the latest Linux operating systems from Red Hat and SUSE. The new offerings will become available starting in the third quarter. Intel also will roll out a new developer community to developers code for the new hardware that will roll out with its HPC technologies, which also include other offerings such as the company's solid-date drives (SSDs), Ethernet products and silicon photonics capabilities. The Modern Code Developer Community will include tools, training and support, and Intel is hoping to reach 400,000 programmers and partners by the end of the year.
The Omni-Path technology is designed to support the faster performance and integrated cores that Xeon Phi brings to the equation. According to Wuischpard, the technology will offer 73 percent better switch messaging rates, 23 percent lower port-to-port latency and 60 percent lower switch fabric latency in clusters than InfiniBand EDR.