ACADEMIA
Microway WhisperStation Silent Personal Supercomputer To Use NVIDIA Tesla GPUs
At Supercomputing 2007, Microway will showcase its new WhisperStation featuring NVIDIA Tesla GPUs. Microway will be providing solutions that utilize both the Tesla C870 GPU computing processor and the Tesla S870 GPU computing Server. These new offerings will have wide acceptance in industries where OpenMP applications employing hundreds of parallel threads are used to speed up floating point operations. These include tomography, oil and gas, simulation of complex molecular systems and other numeric or computationally intensive applications. 
The Tesla C870 processor employs a massively multi-threaded architecture with 128 independent IEEE 754 single precision floating point cores. When coupled with NVIDIA CUDA, a suite of developer tools including a C-compiler, debugger, performance profiler and optimized libraries, one or two Tesla C870 cards can transform a standard workstation into a personal supercomputer, delivering 518 gigaflops of peak floating point performance, 1.5 GB local memory and up to 76.8 GB/sec. memory bandwidth per GPU. The Tesla S870 server incorporates four Tesla GPUs in a 1U platform which can be incorporated into any HPC cluster. Multiple GPUs can be controlled by a single system via the GPU computing driver, delivering incredible throughput on computing applications. The power of the GPU to solve large-scale problems can be multiplied by splitting the problem across multiple GPUs. System monitoring, thermal control and fault notification in the server product provide the necessary features for efficient integration of GPU computing servers into the data center. Benchmarks have shown an incredible increase in GFLOPS by offloading computations to the GPU. VMD, a molecular modeling application from UIUC, has obtained an amazing 291 GFLOPS performing a computation of the coulombic interactions. By comparison, a typical dual Opteron or Xeon system will only get 1-2 GFLOPS. “As multi-core and parallel processing architectures become ubiquitous in high performance computing, computer scientists around the world are beginning to fundamentally advance the pace of their research,” said Andy Keane, general manager of the GPU Computing business at NVIDIA. “The massively multi-threaded core of the GPU positions it at the very heart of this revolution and with the support of companies like Microway, we look forward to making the promise of the ‘personal supercomputer’ a reality. “Historically, math coprocessors have been utilized in the HPC space to provide higher throughput for floating point calculations,” commented Stephen Fried, Microway's CTO. “Starting with the 8087, Microway embraced coprocessor technology to deliver optimal performance for its customers. Now, a new era of coprocessing has begun, leveraging COTS parts to dramatically improve performance while still being very affordable. In fact, utilization of GPU computing improves your price/performance ratio significantly. Power savings can also be realized through reduction in the number of systems needed to effectively compute your applications. It's win-win!”
The Tesla C870 processor employs a massively multi-threaded architecture with 128 independent IEEE 754 single precision floating point cores. When coupled with NVIDIA CUDA, a suite of developer tools including a C-compiler, debugger, performance profiler and optimized libraries, one or two Tesla C870 cards can transform a standard workstation into a personal supercomputer, delivering 518 gigaflops of peak floating point performance, 1.5 GB local memory and up to 76.8 GB/sec. memory bandwidth per GPU. The Tesla S870 server incorporates four Tesla GPUs in a 1U platform which can be incorporated into any HPC cluster. Multiple GPUs can be controlled by a single system via the GPU computing driver, delivering incredible throughput on computing applications. The power of the GPU to solve large-scale problems can be multiplied by splitting the problem across multiple GPUs. System monitoring, thermal control and fault notification in the server product provide the necessary features for efficient integration of GPU computing servers into the data center. Benchmarks have shown an incredible increase in GFLOPS by offloading computations to the GPU. VMD, a molecular modeling application from UIUC, has obtained an amazing 291 GFLOPS performing a computation of the coulombic interactions. By comparison, a typical dual Opteron or Xeon system will only get 1-2 GFLOPS. “As multi-core and parallel processing architectures become ubiquitous in high performance computing, computer scientists around the world are beginning to fundamentally advance the pace of their research,” said Andy Keane, general manager of the GPU Computing business at NVIDIA. “The massively multi-threaded core of the GPU positions it at the very heart of this revolution and with the support of companies like Microway, we look forward to making the promise of the ‘personal supercomputer’ a reality. “Historically, math coprocessors have been utilized in the HPC space to provide higher throughput for floating point calculations,” commented Stephen Fried, Microway's CTO. “Starting with the 8087, Microway embraced coprocessor technology to deliver optimal performance for its customers. Now, a new era of coprocessing has begun, leveraging COTS parts to dramatically improve performance while still being very affordable. In fact, utilization of GPU computing improves your price/performance ratio significantly. Power savings can also be realized through reduction in the number of systems needed to effectively compute your applications. It's win-win!”