News Roundup: The Week in HPC
We’re all fully back from the extended holiday break, or so it would appear given the uptick in the news cycle. To be fair, however, much of what we noticed coming off the wires here in early January had a definite enterprise twist with announcements from SGI, IBM and others about offerings wrapped around large-scale analytics and cloud computing.
On the research front, we saw some notable announcements, including from Indiana University, which will be working on two new supercomputing projects and learned about how the soon-to-be-retired Kraken supercomputer chewed on some complex nanoscale materials problems for researchers at NICS. For those interested in the latter, subscribe to the podcast where the lead scientist on that project talks about the use of Kraken, the evaluation of new architectures for the post-Kraken era, and what his compute-driven research will mean to a number of fields.
On that note, Soundbite listeners will notice that HPCwire is now carrying a daily podcast Monday through Friday featuring interviews with key researchers, thought leaders and news makers in the arena of high performance computing. During our inaugural week of the series we’ve talked with the Ohio Supercomputer Center, a researcher from NERSC, and will continue the trend with more engaging chats in the coming days.
Speaking of research, we noticed that Silicon Mechanics is hosting a research cluster grant competition. They’ve configured an HPC cluster, valued at about $190,000,which includes hardware and software donated by Intel, NVIDIA, HGST, Mellanox Technologies, Supermicro, Seagate, Kingston Technology, Bright Computing, and LSI Logic. This year’s HPC cluster contains eight compute nodes, one head node, Intel Xeon Phi coprocessors, NVIDIA Tesla GPUs, and InfiniBand and Gigabit Ethernet networking. Not bad…more details here.
But back to the top news items for the week ending January 17…We feature an enterprise twist on HPC as most of the announcements that we caught off the wire crossed the HPC and business boundaries. We’ll begin with some positive news for parallel computing on the cheap….
We’ve been following parallel processing startup Adapteva for a number of years here at HPCwire, most recently following their successful Kickstarter campaign, which raised close to a million dollars in community donations. This week more good news came for the company with a $3.6 million investment round from Carmel Ventures and communications technology company, Ericsson.
In its five years, Adapteva has produced a unique, scalable and energy-efficient architecture and IP for parallel processing, delivering two parallel processing accelerator chips in 2011. The 16-core and 64-core Epiphany chips are C/C++ programmable floating-point capable RISC multicore processors able to deliver 50GFLOPS/watt.
Adapteva’s Parallella board, a credit card-sized ARM-based computer with an FPGA and its own Epiphany processor made waves with its $99 price tag, not to mention its performance capabilities for under 5MW.
“Parallel processing and high performance computing (HPC) are of huge interest to us as these markets are developing quickly and changing dramatically,” said Ori Bendori, partner at Carmel Ventures. “Adapteva’s approach is very different than the traditional silicon vendor model and we believe this could lead an industry transformation – just the type of innovation we like to invest in.”
We often share stories about GPU computing in research environments, but a recent announcement this week from Boston Ltd. NVIDIA and Supermicro showed another novel use of acceleration—this time in some new high-flying context.
UK technology company, Aveillant, has successfully completed its two week live demonstrations at Glasgow Prestwick Airport which show that its Holographic Radar reliably removes wind turbines from airport radars with no degradation to radar performance, paving the way for wind farms to be safely built near airports.
Having been in development for two years, the Holographic Radar builds 3D trajectories of moving objects in its field of view and can intelligently differentiate between various types of objects such as aircrafts or wind turbines. Subsequently, it can reliably detect aircrafts and feed their positioning back to the Air Traffic Control system in real-time, allowing controllers to have a consistent and accurate view of the aircraft at all times when passing over a wind farm, maintaining the safety of pilots and the public.
“The processing of these large quantities of data would have previously required around a thousand Cray super computers, costing millions and consuming a substantial amount of energy themselves,” said Peter Wurmsdobler, lead software engineer at Aveillant. “Boston Ltd was able to deliver high performance NVIDIA GPU-powered computers from Supermicro UK, making it both technically and economically viable to continuously process all data captured”.
HP has announced that Air France has automated and increased the reliability of its 1,500 Linux servers by deploying a private cloud solution based on HP Cloud Service Automation (CSA) software to accelerate deployment times for physical and virtual infrastructures.
As HP said today, “With between 350 and 500 installations and reinstallations annually, Air France’s Linux server farm is growing rapidly. In light of the resource costs incurred by these installations, and the increasing convergence between Air France and KLM, it became necessary to standardize on a single tool to automate the deployment of operating systems, middleware, monitoring tools and capacity management. The HP solution, which paid for itself within a year, significantly reduced the time it was taking to bring environments into service while improving quality and reducing operational costs.”
“Since automating our installation processes, we have gone from around six days for installing a server cluster to one day and, instead of more than 24 hours, it now takes less than 15 minutes to install one virtual machine,” said Patrick Bourel, head of Open Systems, Air France. “This scalable platform will allow Air France to put in place monitoring and audit tools in order to achieve better quality of service.”
SGI announced plans this week to develop an in-memory appliance based on the SAP HANA platform. Using the scalable shared memory architecture of SGI’s next-generation UV system together with SAP HANA, the new in-memory appliance from SGI will be designed to streamline database management for single large node environments, which require extremely high capacity and scale to meet the needs of in-memory databases.
According to SGI, “The new SGI computing system is expected to help enable businesses and government agencies running high volume databases and multi-tenant environments to leverage high performance DRAM that can offer up to 200x the performance of flash memory to help deliver faster insight. In addition, SGI’s shared memory technology will enable unprecedented scale in a single node, helping to enable customers to reduce the cost of management by up to 50 percent compared with multi-node solutions.”
“Many companies are running database sizes and multi-tenant environments that require exceptional scale-up capacity capabilities, where SGI leads with its expertise and technology,” said Jorge Titinger, president and CEO of SGI. “By working with SAP, a trusted provider to the enterprise and one of the innovative leaders in in-memory technology, SGI will now be able to target a significant high-end user requirement, representing an exciting market opportunity for us.”
This week IBM announced its new X6 architecture, which applies to its System X and Pure Systems line. The new architecture is aimed at analytics, database and cloud workloads and features integrated eXFlash memory channel storage, a DIMM-based storage offering that provides 12.8 terabytes close to the processor site. IBM says this means that X6 will offer lower latency for database operations and reduce cost overhead by eliminating the need for external SAN/NAS storage units. In addition, they also announced a new storage offering for the cloud, database and analytics market with their IBM FlashSystem 840, which provides 1.1 million IOPS—a doubling in performance over the previous line.
Be sure to check out the events listing to find out where the best places to meet will be this coming year.