Microway customers expect a high level of technical expertise from the sales, integration, and research departments. Our whitepapers serve to engage customers about the latest hardware and software solutions. Please return frequently for new materials.
Next Generation Intel® Microarchitecture "Nehalem"
Intel® usually introduces a new processor every year, alternating between a new microarchitecture and a new process technology, including a die shrink of the current microarchitecture. Nehalem is the latest microarchitecture, using the same 45 nm process used in later versions of the Core microarchitecture (Penryn), but with a brand new design.
Six-Core AMD Opteron Processor Istanbul
In April 2009 AMD announced the first 6-core Opteron processor, codenamed Istanbul. Available for 2-, 4-, and 8-socket systems, with clock speeds ranging from 2.0 to 2.8 GHz, Istanbul is the successor to the quad-core Shanghai processor launched in 2008. It provides up to 30 percent more performance in the same socket and with the same power draw.
InfiniBand Hardware and Fabric Analysis
InfiniBand MPI Network Design and Simulation Using IBSim
By using Microway’s InfiniScope™ real-time InfiniBand monitoring tool to monitor HPC (High Performance Computing) clusters performing real world computations, we have observed that in many HPC applications, the amount of time spent carrying out computations is much greater than the time spent exchanging data between nodes. Microway has created an HPC simulator called IBSim. IBSim gives a network designer the ability to simulate the performance of classes of applications running on an InfiniBand cluster. We demonstrate the use of InfiniScope and IBSim in a case study based on simulation of a two-dimensional nearest neighbor computation model. This paper describes the simulator, presents the study, and explores the results.
FasTree™ InfiniBand Fabrics: A Low Latency Modular Approach to Designing
InfiniBand Fabrics Used in MPI Clusters
Microway has developed a family of InfiniBand FasTree™ switches that work with conventional HCAs. When building large switched networks, the use of FasTree components for leaf nodes and spines has a number of advantages over the popular monolithic "FatTree" design. The advantages include modularity, reduced latency, extensibility, and improved cable management.
GPGPU Architecture Comparison of NVIDIA® and ATI™ GPUs
For customers directly comparing GPUGPU architectures, there have been few detailed technical resources. This whitepaper shows the two main competing GPGPU architectures, their derivative products, and presents Microway solutions optimized for all GPU:CPU ratios and implementations.
GPU Computing: More exciting times for HPC!
In the never-ending quest in the High Performance Computing world for more and more computing speed, there's a new player in the game: the general purpose graphics processing unit, or GPU. Performance speedups between 10x and 1000x have been reported using GPUs on real-world applications. For most people, a performance improvement of even one order of magnitude does not simply increase productivity. It will allow previously impractical projects to be completed in a timely manner.
Microway InfiniScope™ InfiniBand Diagnostic Tool
InfiniScope is a graphical diagnostic tool for InfiniBand clusters. It provides an easy way for customers to vizualize the traffic in their InfiniBand networks, ensure they are running properly, and identify problems like underperforming interconnect links.
Technical Note: Visualization, Measurement, and Improvement of Linux Cluster Performance
Microway provides visualization tools that help identify and address global network and MPI performance issues, and in this note we use the tools to investigate the performance of existing and emerging interconnect technologies.
For personal assistance, please call a sales person who speaks your
language at (508)746-7341. To request a price quote for your next high-end,
fully integrated Microway Opteron Cluster, Server, Workstation, or Storage System go to
the Price Quote Request page.