Hewlett Packard Enterprise (HPE), now about eight months into its transition as a separate entity, retained the prestige of fielding the most systems of any vendor on the Top500 list announced at the ISC2016. HPE had 127 systems (25.4 percent) though the number was down from 155 just six months ago. Other ISC news included introduction of a vertical solution for manufacturing developed with ANSYS, announcement of expanded use of Intel (Knights Landing, Omni-Path) and Mellanox (EDR) technology, and the unveiling of a new HPE software environment for HPC.
To some extent the product announcements represent the end of HPE’s extensive refresh of its HPC line – transitioning earlier SL models into the Apollo portfolio. The news also highlighted HPE’s increasing effort to deliver purpose-built HPC solutions for select markets, said Bill Mannel, VP and GM, High Performance Computing, Big Data and IoT Servers for HPE.
Back in April HPE announced new server offerings including one targeting financial services (an optimized Apollo 2000) and another for deep learning (Apollo 6500), (see HPCwire article, HPE Expands HPC Lineup; Targets Deep Learning, Lustre, and FS). In June the company announced sale of its services business (see HPE Spins Off Services Unit in Merger with CSC). With the latest announcements, HPE’s high performance computing portfolio seems essentially complete.
The new manufacturing solution, said Mannel, “Principally focuses on the small and medium enterprise area. For the most part these customers are using high-end workstations whether one socket or dual. For them, the new solution is a good fit. It sits in a standard rack and provides many HPC capabilities quickly.” The plan is to bring the targeted solution to market via resellers while the CAE software for which it has been optimized is available through ANSYS.
Like many HPC vendors targeting the enterprise HPE’s strategy is to try to ease adoption pain of HPC for that market segment, said Mannel, “We’ve moved for the most part from sort of a semi-custom type of development to looking at market requirements and then designing solutions specifically optimized for HPC and big data.” This approach, it’s hoped, cam help users speed deployment and achieve faster ROI. Expect other vertical solutions from HPE, he said.
HPE also introduced a highly flexible software-defined platform for HPC using the new HPE Core HPC Software Stack with HPE Insight Cluster Management Utility v8.0 (Insight CMV). It’s designed to meet the needs of server cluster environments that may need to scale to thousands of compute nodes.
According to HPE, “The new software stack is a pre-integrated, pre-tested single software suite that combines open source-based application development tools, libraries, and compilers with HPE cluster management capabilities including HPE iLO and simple cluster set up tools. This suite enables developers, IT administrators, engineers and researchers to quickly and easily develop, test, deploy and manage their HPC environments.”
Mannel reports HPE had been supplying the software for a few customers already and was now taking the opportunity to roll it out more generally. At least one user has good things to say about the package. Peter Longreen, COO National Life Science Supercomputer, Technical University of Denmark, Deputy head ELIXIR Denmark is quoted in the press release saying, “HPE Insight CMU has provided us with comprehensive functionality to effectively manage our large cluster environment with rapid bare metal provisioning, simple monitoring with remote management and easy integration with a wide variety of cluster components.”
Interestingly, HPE is an early supporter of OpenHPC, the young effort to create a more standardized HPC software stack. OpenHPC is now officially a Linux Foundation Collaborative Project (see HPCwire article, Heading into ISC16, OpenHPC Releases Latest Stack with 60-plus Packages). Said Mannel, “We’re very involved with OpenHPC and are using it as a baseline for our software environment going forward.”
The other major announcement involved HPE plans for incorporating Intel’s Xeon Phi/KNL processor family and Intel Omni-Path (OPA) technology. The 6000 line will be able to use the new HPE ProLiant XL260a server trays based on the next generation of the Phi and OPA. HPE also beefed up the Apollo 2000 system and ProLiant DL server platforms with Omni-Path fabric options. Mannel said that leveraging Intel’s Scalable System Framework will improve scalability and manageability and permit customers to run HPC applications in a massively parallel manner with minimal code modification.
Amid the ISC din, HPE was hardly alone in trumpeting its adoption of Intel’s newest offerings. Many systems and board makers made similar announcements in what seemed like a well-coordinated show of market traction for Intel technology.
“Customers who have purchased the 6000 with conventional Xeon processors will be able to purchase more chassis and actually have a combined environment,” said Mannel, noting the support for OPA isn’t new. “We had have been supporting OPA for a number of customers but now it’s available throughout the Apollo line.” HPE also announced support for Mellanox enhanced data rate (EDR) devices across the line; EDR had been supported in the 8000 since November.
On balance, HPE does seem committed to the x86 architecture and Intel’s lineup in particular. It has an ARM cartridge for the HPE Moonshot line of datacenter servers, but most of the company’s effort and mindshare have gone to Intel-based Moonshot cartridges. Mannel said candidly, “We continue to build the ARM cartridge, but we also continue to have the most traction with Intel product. We are focusing a lot of our energies around the Intel [Xeon] E3 and [although] we still have interest in other areas, it’s by far our major area of focus for Moonshot.”
Mannel emphasizes HPE remains an enthusiastic user of GPUs where they “make sense.” Deep learning is one example and HPE 6500 deep learning platform makes heavy use of GPUs. “Initially, our interest [there] is in NVIDIA GPUs and we ship all their current GPUs or will be by the end of the summer; we are prioritizing some GPU models over others depending on customer needs. But we offer GPUs pretty much across the entire Apollo line,” he said.
Here’s a snapshot of the new products availability:
- The HPE Core HPC Software Stack with HPE Insight Cluster Management Utility V8.0 is now available for download.
- The HPE ANSYS Solution for CAE is available now through HPE and worldwide channel partners. The CAE software is available through ANSYS.
- HPE Apollo 6000 systems with new HPE ProLiant XL260a server trays will be available in September through HPE and worldwide channel partners.
- The Intel Omni-Path Architecture is now available for initial support on HPE Apollo 6000 and HPE Apollo 2000 systems.
- Mellanox’s EDR 100Gb/s InfiniBand solution is now available for initial support on HPE Apollo 6000 and HPE Apollo 2000 systems.
According to IDC, HPE remains the global leader in HPC market share with 35.9 percent, and IDC analyst Steve Conway offered that “HPE is “upping the ante” with new additions to the HPE Apollo portfolio of purpose-built solutions.” So far the company seems to have avoided costly missteps although selling effectively to the enterprise may prove challenging to all HPC vendors.
It will be interesting to watch how HPE approaches the full HPC market moving forward. As noted earlier, it is the leading vendor in terns of systems currently on the Top500. Like many, Mannel downplays the importance of bragging rights usually attached to placing supercomputers in the Top500 and realistically acknowledges the company’s lack of success penetrating the highest levels of the Top500.
“We’ve tended to be heavily represented in the 100 to 500 versus 1 to 99. I did create over the past year an advanced technology team focused on getting solutions into that top 100, working on technologies involved with HP Labs,” say Mannel.