The Weekly Top Five
The Weekly Top Five features the five biggest HPC stories of the week, condensed for your reading pleasure. This week, we cover the new “Trestles” system at SDSC; Canada’s big supercomputing allocation; NVIDIA’s CUDA Toolkit enhancements; a joint public-private manufacturing initiative; and Supermicro’s latest compact offerings.
SDSC’s “Trestles” Comes Online
“Trestles,” the newest supercomputer at the San Diego Supercomputer Center (SDSC), debuted this week. This 100-teraflop system is available to users of the TeraGrid, the country’s largest open science infrastructure.
Trestles features Appro’s latest quad-socket server outfitted with 8-core AMD Magny-Cours Opterons connected via a QDR InfiniBand fabric. With 324 nodes, the cluster has a total of 10,368 cores. Each server node sports 64 gigabytes (GB) of DDR3 memory and 120 GB of flash memory. On the whole, the system has 20 terabytes of memory and 38 terabytes of flash memory, and runs at a peak speed of 100 teraflop/s. Based on the latest TOP500 list, Trestles would come in at #111.
UCSD and SDSC are pioneering the use of flash technology in HPC systems, a move away from the slower spinning-disk hard drives. Flash got its start in small devices, like mobile phones and laptop computers, but its use in high-end systems is starting to gain momentum. Benefits of flash include faster read/write speeds, higher reliability and better energy-efficiency.
Trestles will serve as a bridge until the more powerful 245-teraflop “Gordon” supercomputer is installed later this year. Gordon is part of a five-year, $20 million NSF-funded project and will employ a large amount of flash memory. ”Dash,” deployed last April, is SDSC’s other flash-based system. All three clusters employ a similar architecture, one that leverages commodity parts in novel ways to maximize performance.
Trestles has been in development since last August, when SDSC announced the $2.8 million NSF award. The supercomputer will be available to TeraGrid users through 2013.
Compute Canada Announces Largest Allocation
Compute Canada, a national platform of advanced computing resources, in partnership with SciNet, Canada’s largest supercomputer center, have announced the largest allocations ever made on Canadian supercomputers, intended to enable huge scientific advances. The high-end resources will be used to boost research in a diverse array of scientific disciplines. Applications include aerospace design, climate modeling, medical imaging, galaxy formation, proton collision and more.
The grants were awarded on a competitive basis, taking into account both the projects’ scientific merit and computational need. It is the role of SciNet and Compute Canada to help Canadian reseachers to create tools and products that improve lives.
Dr. Seth Dworkin, a researcher at the University of Toronto’s Mechanical and Industrial Engineering department, is using the computing resources to study the combustion of biofuels, aiming to develop cleaner burning substitutes for aviation use. Dr. Dworkin states, “The expertise and computational resources at SciNet are helping us tackle problems of combustion-generated emissions using simulations of unprecedented size and accuracy. We’re learning more and more about the formation and nanostructure of atmospheric pollutants and are now able to apply that knowledge to the design of engines and alternative fuels.”
New Manufacturing Effort Relies on HPC
A major manufacturing effort was launched by the Obama administration on Wednesday. The National Digital Engineering and Manufacturing Consortium (NDEMC) was formed to bolster the nation’s small manufacturing enterprises (SMEs) by increasing their access to high-end computing resources. The collection of public-private interests was organized by the Council on Competitiveness in light of poor HPC adoption among small-to-medium-sized manufacturing outfits. Matching underserved manufacturing companies with cutting-edge modeling and simulation tools has been shown to improve product quality, reduce implentation times, and cut costs.
In addition to the Council on Competitiveness, project partners include the Ohio Supercomputing Center, the National Center for Manufacturing Sciences, and Purdue University. Joining them are private partners are Deere & Co., General Electric, Proctor & Gamble, and Lockheed Martin.
Also announced was the Midwest Project for SME-OEM Use of Modeling and Simulation. Part of the greater NDEMC project, the Midwest Project is also engaged in using the power of simulation and modeling to increase manufacturing output.
The US Department of Commerce contributed $2 million in funding with an additional 2.5 million coming from industrial partners. As reported in our feature story, the project is expected to start within the next four to six weeks and last for 18 months. Supporters hope that a successful outcome will lead to to renewed suppport and additional funding.
In related news, the Ohio Supercomputer Center (OSC) has signed a collaboration agreement with the Procter & Gamble Company. The parties will work together on innovative modeling and simulation projects.
NVIDIA Updates CUDA Toolkit
This week NVIDIA announced its latest CUDA Toolkit, version 4.0. The release was designed to enable more developers to take advantage of the parallel programming capability of GPU computing. New features include unified virtual addressing, GPU-to-GPU communication and expanded C++ template libraries.
The company describes the updates as follows:
- NVIDIA GPUDirect 2.0 Technology – Offers support for peer-to-peer communication among GPUs within a single server or workstation. This enables easier and faster multi-GPU programming and application performance.
- Unified Virtual Addressing (UVA) – Provides a single merged-memory address space for the main system memory and the GPU memories, enabling quicker and easier parallel programming.
- Thrust C++ Template Performance Primitives Libraries – Provides a collection of powerful open source C++ parallel algorithms and data structures that ease programming for C++ developers. With Thrust, routines such as parallel sorting are 5X to 100X faster than with Standard Template Library (STL) and Threading Building Blocks (TBB).
A release candidate of CUDA Toolkit 4.0 is available starting this Friday.
For more in-depth analysis, check out our feature coverage.
Supermicro Debuts 8-Way Server
Supermicro introduced its 8-Way Enterprise Server at the CeBIT trade show in Hannover, Germany, on Monday. The solution comes with up to sixty-four Xeon processor cores packed into a 5U enclosure that also includes 64 DIMMs, up to 10 PCI 2.0 expansion slots, and 24 2.5″ hard drives. An 80-core option is coming soon, according to the company. Supermicro also launched its SuperBlade system, which supports 20 GPUs in a single 7U blade.
Charles Liang, CEO and president of Supermicro, reported on the new offerings:
Compared to other GPU-enabled blade solutions Supermicro’s GPU SuperBlade provides more than double the number of GPUs per 1U of rack space, and our 8-Way SuperServer is unique in the industry with support for eight next-generation ten-core Intel Xeon MP processors in a 5U form factor, up to 2TB of memory and 10 PCI-E 2.0 slots for high availability and virtualization support. Taken together with Supermicro’s broad line of 1U, 2U and 4U GPU supercomputing servers, we have established undisputed leadership in the new and rapidly evolving GPU computing business space. These new products play well into our strategy to deliver the end-to-end IT hardware needs of datacenter, HPC and server farm customers.