HPE shared memory high performance computing
Tackle complex, data-intensive HPC problems holistically with the unique scale-up architecture of HPE Superdome Flex family
+ show more
Organizations across the globe utilize high-performance computing (HPC) to solve difficult problems in science, engineering, and business.
Many depend on the HPE ecosystem of accessible, affordable HPC solutions to gain the computational power they need and to accelerate time to discovery. And as HPC problems grow in size and diversity, HPE solutions equip research, development, and security teams to meet current and future requirements.
HPC workloads are commonly run on clustered systems, in which computational problems can be distributed across multiple servers (nodes) working in parallel and which are connected over a high-speed network with shared storage. In addition to extending computational capabilities by orders of magnitude, HPC teams can run many types and sizes of jobs within and across nodes concurrently.
Some HPC problems, however, are challenging to run across multiple nodes.Complex and often data-intensive with strong interdependence, these problems are often best tackled by one “shared-memory” node, known as a “fat” node, rather than breaking jobs into pieces across many small nodes. Examples of such workloads include:
- Computer-aided engineering (CAE) such as optimizing placement of airplane antennae and GPS devices to minimize interference using electromagnetic simulation, improving the structural design of wind turbines using FEM 1 simulation, designing screw augers for transporting bulk material using DEM 2 technology, or perfecting race car designs using computational fluid dynamics to simulate aerodynamics.
- Medical and agricultural genomics such as genome mapping, a process that compares billions of small sequences and terabytes of data with a previously assembled genome until complete, genomic research in which a full DNA sequence is analyzed to identify gene functionality and variations to help predict, diagnose and treat disease, or precision medicine, an approach that considers individual gene variabilities to provide personalized patient treatment.
- Fraud prevention, such as scanning and comparing barcodes with previously scanned barcodes (thousands to billions) to identify copies before loss or damage is incurred.
- In situ visualization, in which a simulation is run and the data generated is analyzed visually and in real time.
Accommodating such workloads on the largest nodes can be challenging for HPC
teams as they often compete for time, leaving some jobs waiting and less-powerful nodes underutilized. This increases time and effort, and reduces efficiency.
Such workloads might also be too big for the node to handle. If the node’s memory is exhausted, the job will fail, wasting hours of running time. To fit on the node, models, and simulations must then be made smaller, for example by using sections and estimations or decreasing granularity. This, however, reduces accuracy, increases prototype costs, and adds time to production or discovery.
An additional challenge for relatively small HPC teams is the question of who will manage the cluster environment, a role that includes balancing workloads, tuning performance, and other IT tasks. Can the IT department dedicate a cluster administrator, or must an engineer or scientist take time from development and research to learn and perform?
Solution: shared memory HPC with HPE Superdome Flex family
The ideal solution for solving complex, data-intensive problems holistically is with HPE shared memory high performance computing and the breakthrough SMP (symmetric multiprocessing) server family of HPE Superdome Flex and HPE Superdome Flex 280 servers.
Providing unparalleled scale‑up compute and shared memory resources with single‑system simplicity, HPE Superdome Flex family equips your HPC team to:
- Accelerate time to discovery as jobs no longer compete and wait for “fat” nodes, enabling more jobs to be completed in less time.
- Improve accuracy using simulations with larger models and more data.
- Increase productivity by freeing scientists and engineers from cluster management tasks.
Software developers will also greatly benefit from HPE shared memory HPC solutions because, unlike when moving from a desktop to a cluster, there is no need to modify code. HPE Superdome Flex and HPE Superdome Flex 280 servers are like giant Linux® workstations with lightning speed.
This server family delivers:
- Scale from small to huge. HPE Superdome Flex family unique modular architecture allows you to start small and grow at your own pace without sacrificing performance. HPE Superdome Flex 280 offers a low entry point and granular scaling, starting at 2 and scaling as a single-system to 8 sockets in two-socket increments. The server is designed to provide 64 GB to 24 TB of shared memory using DRAM or in combination with persistent memory. HPE Superdome Flex scales from 4 to 32 sockets in 4-socket increments, with shared memory scalable from 768 GB to 48 TB.
- Optimum flexibility. HPE Superdome Flex family modularity and scale helps you avoid overprovisioning, disruptive upgrades, and the cost and complexity those carry. Shared memory capacity is provided using DRAM only, or a combination of DRAM and Intel® Optane™ Persistent Memory for HPE. You can add NVIDIA® GPUs if your workload requires this technology.
- Unbounded I/O. Capitalize on a well-balanced I/O subsystem for high performance. Both models, HPE Superdome Flex 280 and HPE Superdome Flex, offer two I/O choices: A 16-slot or a 12-slot, per 4-socket chassis, and support a multitude of cards providing rich solution flexibility.
- Extreme availability. Unique HPE RAS features delivered in the HPE Superdome Flex family include Firmware First, which contains errors at the firmware level and insulates the OS; a built-in Analysis Engine to monitor, analyze, and fix errors with self-repair; and advanced resiliency across every subsystem. For maximum application availability you can add HPE’s leading clustering solution HPE Serviceguard for Linux. And on Superdome Flex 280, silicon root of trust protection and support for Trusted Platform Module (TPM) 2.0 further protect data. For details on all HPE Superdome Flex family RAS capabilities, check these white papers: HPE Superdome Flex 280 | HPE Superdome Flex.
- Simplified user experience. Leverage HPE OneView Management, OpenStack cluster management, industry-standard Redfish APIs, and in the HPE Superdome Flex 280 model, a simplified management GUI. You can also consume HPE Superdome Flex family as a service, via HPE GreenLake. Learn more about HPE Superdome Flex management in these white papers: HPE Superdome Flex 280 | HPE Superdome Flex.
Organizations from a wide range of industries are leveraging HPE Superdome Flex to tackle data-intensive HPC workloads holistically.
Czech Technical University: Powered-up research capacity
Founded in 1707, Czech Technical University is one of the world’s oldest technical universities. With HPE Superdome Flex, scientists at CTU’s Faculty of Information Technology now experience almost unlimited computing power to research machine learning, robotics, computer vision, data mining, and more.
“Superdome Flex is very scalable with the lowest deployment time, and has the capabilities to meet our existing and future needs”
– Martin Vaňko, Head of ICT Services, Czech Technical University
NIG: Building a genomic supercomputer
The National Institute of Genetics is the center of life science and genomic medicine research in Japan, and provides researchers with opportunities for joint use of and research with supercomputers. NIG adopted Superdome Flex as its “5th generation supercomputer” to manage large-scale and complex environments in an integrated manner and respond quickly and efficiently to researchers’ needs.
“A large shared memory space exceeding 10 TB is very effective for assembly processing that connects a large number of genome fragments to restore sequences”
– Dr Osamu Ogasawara, Project Associate Professor, NIG
Institute of Theoretical Chemistry at The University of Vienna
The Institute of Theoretical Chemistry at The University of Vienna is leveraging HPE HPC solutions, including HPE Superdome Flex, in research simulations around photochemistry—how molecules interact with light.
“The Superdome Flex gives us a favorable CPU cost for one machine with a great deal of memory. It’s robust, cost-effective, and flexible, so we often run 10–20 small simulations concurrently.”
– Markus Oppel, Senior Scientist, Institute of Theoretical Chemistry at The University of Vienna
COSMOS: Revealing the Big Bang
The UK’s COSMOS advanced computing facility, co-founded by Stephen Hawking, runs complex simulations and real-time analyses of terascale datasets. The Superdome Flex computing platform is opening up new research horizons, helping COSMOS achieve its objectives of developing a seamless history of the Big Bang and understanding the gravitational waves of black holes.
“In-memory computing allows us to ingest all of this data and act on it immediately, trying out new ideas, new algorithms. It accelerates time to solution and equips us with a powerful tool to probe the big questions about the origin of our universe”
– Paul Shellard, Professor of Cosmology and Coordinator of COSMOS
Go further, faster, with HPE High Performance Computing solutions
HPE is the global leader in high performance computing solutions, with deep expertise across HPC workloads and a powerful, purpose-built product portfolio that make supercomputing more accessible and affordable for organizations of all sizes.
Ask your Hewlett Packard Enterprise sales representative about equipping your team with shared memory HPC solutions with HPE Superdome Flex 280 and HPE Superdome Flex.
Intel Optane and Intel Xeon are trademarks of Intel Corporation or its subsidiaries in the U.S. and/or other countries. Linux is the registered trademark of Linus Torvalds in the U.S. and other countries. NVIDIA is a trademark and/or registered trademark of NVIDIA Corporation in the U.S. and other countries. All third-party marks are property of their respective owners.
- 1 Finite Element Method
- 2 Discrete Element Method