Microway HPC Times

Visit us at Supercomputing 2009 in Booth 434. Microway will be demonstrating more powerful solutions for high-throughput computing with GPUs, multicore CPUs, InfiniBand, advanced server cooling, and cluster software tools. Microway-engineered InfiniBand products include 36-port QDR FasTree Switches, ConnectX TriCom HCAs with 1 microsecond latency, MPI Link-Checker network performance validation and InfiniScope real-time InfiniBand analysis and diagnostic software. We will also be in AMD's Booth #1417 showing Fiorano products with Six-Core Opteron processors and a 2U server with four ATI GPUs. Register for a FREE SC09 Expo Pass

Microway Product Announcements

Dense Computing with NumberSmasher™ 2U Twin². This elegant design includes four compute nodes in 2U with hot-swap motherboard modules, hot-swap hard drives and optional redundant power supplies. A clean cableless design and high-efficiency cooling and power supplies make the Twin² ideal for greener HPC. Available with Xeon and Opteron CPUs.

Maximize GPU/CPU ratio with OctoPuter™-Tesla 4U System! Our 4U system includes 8 NVIDIA Tesla C1060 GPUs and 2 Quad-Core Intel^® Xeon Nehalem CPUs— achieving 8 TFLOPS. This enables a new level of performance for complex modeling, simulations and analysis. Get 16 TFLOPS in a 9U cabinet, as shown at left.
24 Opteron cores in one server with Navion! Navion 4 Way Opteron Server includes a quad-socket motherboard based on the latest AMD SR5690 chipset, utilizing four Six-Core AMD Opterons. Full support for HT Assist technology enables a memory transfer rate of 42GB/s. These servers are offered in 1U, 2U and 4U designs.

FasTree™ 36-port QDR InfiniBand Switches provide outstanding interconnect performance. Their modular, expandable design provides lower latencies than other manufacturers' switches. FasTree is a cost-effective solution for both small and large fabrics.

Microway Customer Projects

Microway HPC customers are taking advantage of our GPU compute clusters based on the Six-Core AMD Opteron and Quad-Core Intel Xeon 5500 system designs. Custom configurations have been delivered recently to many institutions including Harvard University, Texas A&M University, Johns Hopkins University, University of Arizona, Northwestern University, Georgia Institute of Technology, and US government research labs.
GPU Developers are improving application performance with our WhisperStation-Tesla PSC. For modest budgets, a WhisperStation with up to four GPUs can provide more TFLOPS than a traditional compute cluster. For example, AMBER on a Tesla GPU runs 9 times faster than a dual-socket Quad-Core Xeon node. Optimize your GPU code on WhisperStation; then scale up to a custom-configured Tesla GPU cluster.

Trade Show Recap

NVIDIA GPU Technology Conference (GTC): Microway demonstrated the WhisperStation-Tesla PSC for GPU development. A single Tesla GPU can provide up to 1000x speedup vs standard processors.
HPC on Wall Street: Microway's booth showcased NVIDIA's Tesla S1070 GPU compute servers for use in the financial markets. Our WhisperStation-Tesla PSC accelerated MATLAB applications using GPUs with AccelerEyes Jacket.

GPU Programming Updates

Online Resources: There are many technical resources for learning more about GPU computing. For example, Seismic Processing with CUDA.
OpenCL downloads are now available from NVIDIA and AMD.

Visit Terri, Brett, Rick, Mike, Steve and Ann in Portland at Booth #434. See our 4U GPU server with 2 Intel Xeon processors and 8 Teslas and other products mentioned above. We all look forward to visiting with you in Portland.

Microway® HPC Times - October 2009

Microway^® HPC Times - October 2009