Datanami
EnterpriseAI
HPCwire Japan
QCwire
HPC & AI Wall StreetSC15 is starting to feel like EnterpriseHPC15 — not that there isn’t a lot of extreme scale computing on the menu — but the big systems and chip makers all seem to be publically pivoting towards the enterprise. Yesterday it was Dell’s turn declaring its intention to democratize HPC and accelerate mainstream adoption.
Dell, which bills itself as the lone desktop to petaflops company now that HP has split up, is rolling out what it calls engineered solutions and reference architectures intended to make HPC adoption dramatically easier for small and medium enterprises. Engineered solutions can be ready-to-run out of the box, said Jim Ganthier, VP and general manager, Engineered Solutions and Cloud, Dell. The reference architectures are more likely to deliver 80 percent of the solution with Dell-client collaboration filling in the final 20 percent.
Put simply, the Dell belief is this: HPC is becoming increasingly critical to how organizations of all sizes innovate and compete. Many organizations lack the in-house expertise to configure, build and deploy an HPC system without losing focus on their core science, engineering and analytic missions.
Indeed every big systems builder seems to have a version of the argument. Ganthier cited National Center for Manufacturing Sciences data that suggests 98 percent of all products will be designed digitally by 2020, yet 95 percent of the center’s 300,000 manufacturing companies have little or no HPC expertise.
If these estimates are even modestly accurate, it’s no mystery why HPC product makers are shifting course. One could argue that in recent years HPC technology suppliers have mostly wrangled over market share while the total size and growth rate of the traditional HPC market remained modest and driven (mainly) by government procurement cycles.
Moreover, if supercomputers are the spearhead of advanced computing, the slowdown in Moore’s Law has flattened the spear tip just as the enterprise’s need for advanced computing has grown. Result: Some of the grandeur has slipped from the TOP500 and making the list is not the badge of glory it once was. Ganthier said, “It’s kind of starting to run its course because standing up something [in the TOP500] is just a beauty contest.” The real action, Dell is betting, is putting HPC to work. He said Dell wouldn’t walk away from the TOP500 – there were 13 Dells on the list this year – but it wasn’t a priority.
Dell plans to leverage its technical expertise, extensive work with prominent domain collaborators, and alliance with leading technology partners to create a pipeline of pre-formed HPC solutions. A good example is the fruit of its collaboration with genomics powerhouse, the Translational Genomics Institute. The Dell HPC System for Genomic Data Analysis is designed to meet the needs of genomic research organizations to enable cost-effective bioinformatics centers delivering results and identifying treatments in clinically relevant timeframes while maintaining compliance and protecting confidential data.
The engineered solution version is “literally a part number. Everything is prebuilt and prewired right up to including genomics analysis tool suites,” said Ganthier. A more sophisticated user might choose to work with the reference architecture so as to create a system optimized for specific needs.
Two other application specific HPC offerings announced include:
- Dell HPC System for Manufacturing is designed for customers running complex manufacturing design simulations using workstations, clusters or both. Applicable use cases include Finite Element Analysis for structural analysis using ANSYS Mechanical & Computational Fluid Dynamics for predicting fluid behavior in designs utilizing ANSYS Fluent or CD-adapco STAR-CCM+.
- Dell HPC System for Research is designed as a foundation, or reference architecture, for baseline research systems and numerous applications involving complex scientific analysis. This standard cluster configuration can be used as a starting point for Dell’s customers and systems engineers to quickly develop research systems that match the unique needs of research customers requiring systems for a wide variety of research agendas.
“HPC is no longer a tool only for the most sophisticated researchers. We’re taking what we’ve learned from working with some of the most advanced, sophisticated universities and research institutions and customizing that for delivery to mainstream enterprises,” said Ganthier.
The new Dell HPC System Portfolio is a family of HPC and data analytics solutions featuring: simplified design, configuration, and ordering in a matter of hours instead of weeks; domain-specific design that’s designed and tuned by Dell engineers and domain experts for specific science, engineering and analytics workloads using flexible industry-standard building blocks; and fully tested and validated with a single point of hardware support and a wide range of additional service options.
Big data and the cloud considerations were also important if not the major drivers in the strategy shift. “Every conversation about HPC now has a big data component or has a cloud component,” said Ganthier. Data handling and data analytics will no doubt play important roles in product development, as will efforts to leverage Dell cloud resources.
Investment and realignment of resources are needed to make the new strategy work. Dell announced a new expansion of its Dell HPC Innovation Lab in cooperation with Intel specifically for support of its Intel Scalable System Framework. It’s a multi-million dollar expansion to the Austin, TX, facility and includes additional domain expertise, infrastructure and technologists. It’s likely more such centers will be added over time.
Ganthier noted Dell is an early supporter of the nascent OpenHPC project being organized by the Linux Foundation. Its primary goal is a more ‘standardized’ software stack for HPC. It’s still early, but there’s agreement that such a stack that was reliable and easy to use could accelerate HPC adoption in the enterprise.
Dell is clearly betting big on Intel. Beyond becoming the first major original equipment manufacturer (OEM) to join the Intel Fabric Builders program, Dell is working closely with Intel to support its Intel Scalable System Framework, which includes Intel Omni-Path Fabric technology, next-generation Intel Xeon processors, the Intel Xeon Phi processor family, and the Intel Enterprise Edition for Lustre.
Dell also announced its new networking H-Series switches and adapters for PowerEdge servers would support Omni Path. The architecture includes advanced features such as traffic flow optimization, packet integrity protection and dynamic lane scaling allowing for finer-grained control on the fabric level to enable high resiliency, high performance and optimized traffic movement.
In addition, Dell said it would showcase at SC15 many components of the Intel Scalable System Framework including Intel Omni-Path Architecture, Intel Enterprise Edition of Lustre, and Intel Xeon Phi processor family. In addition, Dell is hosting numerous confidential advisory sessions with customers at the show gathering insights to help optimize its implementation of systems using next-generation Intel Xeon Phi.
Mellanox, a longtime Dell collaborator, announced additional investment in Dell’s existing HPC Innovation Lab to provide an end-to-end EDR 100Gb/s InfiniBand supercomputer system. The system is designed to showcase extreme scalability by leveraging the offloading capabilities and advanced acceleration engines of the Mellanox interconnect as well as provide application specific benchmarking, and characterizations for customers and partners.
It’s also worth noting Dell still has its fingers in the big systems pot, often in conjunction with the convergence of big data and cloud initiatives. For example, the San Diego Supercomputer Center (SDSC) recently launched its new Comet petascale supercomputer powered by Dell PowerEdge C6320 servers. “Comet provides ‘HPC for the 99 percent’—serving as a gateway to discovery for a much larger research community so we needed a solid hardware foundation,” said Michael Norman, SDSC director and principal investigator for the Comet project.
Other impressive Dell installations include:
- The Texas Advanced Computing Center at the University of Texas at Austin, which specializes in scientific visualization, data analysis and storage systems, software, research and development and portal interfaces. Dell technology is used in TACC supercomputing clusters, including Stampede and Wrangler.
- The HiPerGator at the University of Florida is now being expanded to add capacity and capabilities with 30,000 cores in approximately 1,000 nodes made by Dell. HiPerGator performs complex calculations and data analyses for researchers to find life-saving drugs, make decades-long weather forecasts and improve armor for troops.
- Jetstream, with Dell hardware at the core, is scheduled to enter production in January 2016 at Indiana University. “[It] will bring to XSEDE and the national research community a user-friendly cloud environment that allows researchers to analyze their data now – whenever now is for that researcher,” said Craig Stewart, Executive Director, Indiana University Pervasive Technology Institute and Associate Dean, Research Technologies, Indiana University.
