At the International Supercomputing Conference in Frankfurt today Intel showed off a bevy of forthcoming additions to its high-performance computing portfolio, including for the first time a switch built around the Intel Omni-Path Architecture 100 Series processor it claims is capable of delivering 195 million messages per port.
In addition, Intel provided updates on how second-generation Intel Xeon Phi processors, codenamed Knights Landing, will be used to provide parallel computing capabilities inside the data center, while also revealing that it has allied with HP to establish new centers of excellence for HPC in Frankfurt and Houston.
Charlie Wuischpard, vice president of the Intel Data Center Group and general manager of Workstations and HPC at the company, said Intel is trying to rally developers around these next-generation supercomputer platforms. To that end, the company expects 400,000 developers and partners with tools, trainings and support to be in place by the end of the year, with 10,000 of them having remote access to both Xeon and Xeon Phi systems to enable them to build applications.
Intel expects to see its OEM partners delivering new systems based on the latest Xeon Phi processors late this year. Meanwhile, switches based on the Omni-Path 100 series are expected to deliver port-to-port latency as low as 100 to 110 nanoseconds, which is 23 percent lower than InfniBand switches at what Intel says will be a much lower cost.
The biggest challenge facing the chip giant with Phi-powered HPC servers will be educating developers on how to create applications that take advantage of the processors to execute code in parallel. To facilitate that process, Intel will also provide developers that build the best applications to help a social cause with a free trip to the CERN supercomputer facility in Switzerland, said Wuischpard.
“We’re planning to reach out to them via a number of hackathons where they can show off their skills,” he said. “We also think in aggregate that developers will see Intel providing the best total cost of ownership.”
As part of that TCO focus, Intel is also showcasing today the latest cloud technologies and editions of Intel Lustre storage software that are due out in the third quarter. The company also said that version 3.0 of Lustre Enterprise Edition, due out in 2016, will not only be faster but will also provide enhanced security and manageability using snapshots enabled by ZFS.
Intel is especially anxious to increase the size of its footprint in the HPC server market, which IDC expects to reach a total of $15.2 billion in 2019. The HPC category is expected to grow significantly faster than the rest of the server market.
While Intel-based systems lower the cost of building HPC applications, for Intel they clearly represent a higher-margin opportunity than traditional data center servers.