It’s difficult to tell what kind of year lies ahead on the supercomputing systems sales side with so much riding on the upcoming release of key Intel products, particularly the new Knight’s Landing chips, among others in the future Xeon line. While the ecosystem might have been a bit more diverse a few years ago, when it comes to the downtime in advance of major architecture shifts, the mighty chipmaker is driving the train—and the rest of the cars aren’t moving until it puts the machine in gear.
This waiting game applies to several system vendors catering to HPC, including Bull, as we noted earlier While the big government-fed supercomputing and large-scale commercial HPC procurements for oil and gas companies might be reduced to a trickle for current generation systems, Cray too is expecting a boom in the coming years—one that can be heard far beyond the halls of supercomputing.
Even though the system vendors and large centers alike seem to be on hold waiting for a new rush of choices, some have already announced new machines to sport the future Intel products, including NERSC with its Cori supercomputer—one that will feature the next-generation Knight’s Landing cards with on-package memory. This Cray-built machine will be one of many we can expect coming in the new wave of procurements down the line, said Cray CEO, Pete Ungaro during a sit-down with us at ISC.
Part of what gives Ungaro comfort is that their entire business isn’t riding on the massive supercomputers that drove over 90% of their business in the past. He says that while they’re still a supercomputer company with unshakable roots in HPC, they’re seeing an unprecedented opportunity to push farther past that affiliation given the data-driven demands of enterprise users in markets that might never call what they do HPC—and which might not be, if applications are definitive. In the last two years they’ve bolstered their storage profile, added “big data” products to the mix, including a graph analytics appliance, and diversified the capabilities of their systems to span a wider area of enterprise workloads.
Last year at ISC he told us that around 10% of Cray’s business was commercial-this year, he expects that number to inch closer to the 20% mark, leaving them less dependent on the slings and arrows of product delays for supercomputing-specific processors (Knight’s Landing again the example) and more capable of filling any gaps with new outreach into enterprise.
The fact is, says Ungaro, more users across a wider swath of business aren’t getting what they need from vanilla commercial systems and they’re exploring options leverage some creative systems engineering to enable their work on both the data and computational horsepower fronts. In addition to focusing on their HPC bread-and-butter business, their targets in the coming year are around, you guessed it, big data via their storage and graph appliance product lines as well as striving toward “openness” via a few key moves through support of the OpenStack foundation, among other things.
While he agreed that Cray’s first quarter results don’t necessarily reflect the growth he described, he cautioned us to sit tight for some big announcements on both the procurements and product fronts this coming year. And besides, Cray, like many companies in the HPC systems business, isn’t one to be judged on a quarter-by-quarter basis since large procurements and other acts of funding can skew results, just as we’re seeing in recent times. After all, what major procurement cycle wouldn’t push farther out to wait for what’s coming when it promises more compute, efficient and I/O than what’s current available. While this makes sense, there aren’t any formal official dates for the launch of new Intel HPC products, although we’re pretty well aware this is later 2015, perhaps into 2016.
Even if the most recent numbers aren’t reflecting it, Cray is well positioned as a company with strong engineering, its interconnect and software ecosystems, and an increasingly diverse customer base. And of course, not everyone at the large-scale supercomputing level is sitting around waiting on the sparklier, newer things. There are plenty of users who have no need to wait for the next big thing from Intel. For instance, at ISC Cray announced a $54 million contract to provide the Korean Meteorological Administration two next-generation “Cascade” Cray XC systems coupled with their Sonexion storage. Most major weather forecasting centers buy pairs of systems, both because of the mission-critical nature of their operations as well as to test and run other applications in addition to the weather models. You might recall a discussion last year on this topic with Cray CEO Pete Ungaro and Isabella Weger from ECMWF—another major weather modeling and forecasting user of the Cascade systems.
They’ve sold other big systems this year, including an XC30 to the Hong Kong Santorium and Hospital for gene sequencing and life sciences research. Additionally, they’re seeing solid business on the storage front, evidenced by the adoption of the Cray Tiered Adaptive Storage (TAS) at the North German Supercomputing Alliance. With the growth on the storage side, coupled with the recent news that they’re hooking Lustre into TAS, they seem to be looking to continue the fine balance between big data storage and management needs with the HPC engineering bent they’re known for.
Notice that aside from the Cori system, these are Asian supercomputing deals, which adds light to Ungaro’s statement that they’re expanding their worldwide footprint with new centers around the globe. While we didn’t get a sense of when these would open, Europe and Asia are new targets—and places where they have key installations already at weather and research centers (and likely commercial operations they’re not allowed to discuss).
In other words, Cray and others should have big procurement news around Supercomputing and a new slew of them at next year’s ISC—and then things get interesting again. By mid-2016, into 2016, the hope is few options for processors will broaden and a new competitive (read as more interesting and diverse) ecosystem will emerge. OpenPower is promising. ARM is promising. AMD will probably do…something. NVIDIA with its IBM partnership and continued growth around CUDA and its own accelerators will continue to be a force. The ecosystem isn’t “on hold” since not everyone is waiting on the shiniest new toys, but it’s going to be quieter on the procurement side for a while—just as it has been SC last year.