While there may not be a distinct, publicly available roadmap laying out the exact course for the next generation of Xeon Phi, conversations are brewing about how smooth the highway to Knight’s Landing will be, what the destination will look like, and of course, exactly how long it will take to arrive.
The unofficial landing of the newest member of the Knights family is expected sometime early next year and although we tried, it was hard to pry more detail from the HPC heads at Intel we spoke with today. We were, however, able to glean some information about some recent upticks in investments around both the existing and future Xeon Phi from one of the newest members of the high performance computing team.
Today marked the first time we’d talked with Charles Wuischpard about the Intel HPC roadmap since he took the position in January following a seven-year stretch at Penguin Computing as CEO. When we profiled Wuishpard following his listing in our annual People to Watch for 2014, he told us that leaving Penguin was a difficult decision. As he noted, “I was quite happy at Penguin Computing. We were coming off another record year and not looking for a change, but when Intel called, I had to listen. What drew me to the opportunity was the chance to be involved in, and participate in the future of HPC at the highest levels from the global race to exascale computing to the expansion and adoption of new HPC computing models like cloud. The assets at Intel to help make this happen are impressive.”
Wuischpard is certainly not a newcomer to technical computing. He spent 17 years at IBM in various executive sales roles in addition to shorter term stints at various other companies, including Versant and GoldenGate Software before taking over as CEO of Penguin Computing in 2007.
In his new position as VP and General Manager of the Datacenter Group’s workstations and HPC division, he reports to Raj Hazra, who will continue to serve as VP and General Manager of Intel’s Technical Computing group with an expanded role in pushing the Phi line (and application readiness goals) into more environments. While Wuishpard admitted that it may seem like there is a great deal of overlap between these two roles, the two are on complementary but divergent trajectories toward the same ultimate mission—to “bring the world to the edge of Moore’s Law.”
Among the technologies on Intel’s horizon that Wuishpard says are critical to the long-term success, the Xeon Phi is at the top of the list, both as it stands now and where he expects to watch it migrate in the future. He added that another key area for investment is in the future of high performance computing for clouds, noting that Intel is emphasizing the importance of cloud-based delivery of HPC with numerous cloud vendors that need to deliver more than just vanilla boxes for demanding workloads. He also pointed to another aspect of Intel’s business that he expects to add to growth–the TrueScale fabric. “I’ve always been a fan of QLogic, even before the Intel acquisition,” he told us, noting that TrueScale’s capabilities are one of the great untold stories along the manycore path.
All of Wuishpard’s excitement is backed by increased investment on the technical computing side, according to Harzra. The high performance computing division’s emphasis along the Knight’s line is focused on engineering, as always, as well as on making sure the application environments to meet the new era of manycore computing are primed. This means extensive work inside Intel and at their Parallel Computing Centers of Excellence (more of these are expected to be announced this year) where they’ll tailor ISVs community, and custom codes for industry.
“As we move from one level of parallelism to the next—from the multicore to the manycore era—we see [Knight’s Landing] as the basic compute architecture for far more than technical computing workloads, especially since adding more parallelism is the only way to get more performance. This will be the compute basis for high performance and compute-intensive applications going forward,” said Hazra.
Hazra and Wuischpard both agree that the current reach of Xeon Phi in terms of messaging, community engagement and overall breadth of knowledge isn’t as long as they might like it to be, given that the emphasis on the product’s development and use is centered on pure HPC workloads and environments. However, both Intel execs noted that this is expected to change rapidly, particularly in the wake Knight’s Landing and future developments. Hazra pointed to the many investments Intel is making internally, including Wuischpard, whom he says will bring an entrepreneurial approach to the Knights family.
Wuischpard told us that he’s been in conversations already with the top five Intel customers, including CERN, about what their future needs are system and code-wise for the 2020-2023 timeframe. He remarked, “Twenty years ago, they were planning for the CERN installation and design of the codes with the assumption that they were going to have a single core 8GHz processor. So that is what they wrote their code for, but now, how do they write their codes to take advantage of a manycore future?”