Hewlett Packard Enterprise (HPE) will be purchasing Cray to the tune of $ 1.3B. The deal represents a 17 percent premium over Cray’s current stock price. Cray, of course, is Cray — one of the leading companies in supercomputing, though its own role in that process has transformed over the years. In the beginning, Cray designed its own, fully custom hardware and software stacks. Today, the company functions as a high-end integrator, working with CPUs from AMD or Intel, but adding its own expertise in interconnect, software, and I/O technology to create the supercomputer stack.
Cray is still a major player in the US supercomputer market. It has a deal with the DoE to provide a supercomputer named Frontier to Oak Ridge Laboratories by 2021 and it’s a subcontractor firm on the 2021 Aurora supercomputer as well. Both of these machines are expected to be exascale-class and represent the first two exascale systems the United States is expected to purchase. HPE will now own core components of the exascale ecosystem in the US, including Cray’s Shasta system architecture and its Slingshot interconnect. HPE has its own work in the exascale and HPC arenas, but it doesn’t have Cray’s name recognition or equivalent deep focus in the market.
Cray and HPE expect to deliver the following benefits from their merger:
- Future HPC-as-a-Service and AI / ML analytics through HPE GreenLake
- A comprehensive end-to-end portfolio of HPC infrastructure – compute, high-performance storage, system interconnects, software and services supplementing existing HPE capabilities to address the full spectrum of customers’ data-intensive needs
- Differentiated next-generation technology addressing data-intensive workloads
- Increased innovation and technological leadership from leveraging greater scale, combined talent and expanded technology capabilities
- Enhanced supply chain capabilities leveraging US-based manufacturing
Greenlake is HPE’s Hybrid Cloud service. The rest of this basically emphasizes that HPE wants to leverage Cray’s expertise in exascale supercomputing design to bring some of that knowledge into its own product families. There’s also talk of extending the concept of HPC-as-a-service, though it’s not clear how large that market is yet. Offering HPC instances for data processing and number crunching would be an incredible boost for labs that could benefit from these capabilities, but can’t afford to purchase the necessary hardware — but much of the performance in HPC is also thanks to careful software tuning, at a level that’s rare to find in the regular software market. It isn’t clear whether these sorts of benefits would transfer to HPCaaS in the same fashion.
HPE’s deal with Cray also illustrates how much additional expertise, beyond CPUs and GPU designs, go into the construction of exascale and HPC systems. Creating solutions that sell into this space is quite difficult. At scale, large computer installations face particular challenges related to keeping workloads synchronized, taking full advantage of the underlying hardware, and making certain power, cooling, and other resources are distributed to where they need to go, minimizing bottlenecks without blowing up costs. Even with decades of its own work in this space, HPE felt it was wiser to buy a company like Cray than attempt to reinvent the wheel. With AI accelerators, machine learning, and more additional silicon being integrated into platforms alongside conventional chips, Cray’s ability to build interconnects and heterogeneous systems architecture is more important than ever.