AMD, Nvidia launch GPUs at number-crunching scientific, engineering apps

Ann Steffora Mutschler, Senior Editor -- 6/16/2008

Specifically designed to accelerate critical algorithms in high-performance computing (HPC), mainstream and consumer applications, microprocessor maker Advanced Micro Devices today detailed the FireStream 9250 stream processor.

To comment on this news see, "Graphics chips challenge x86, Cell."

Separately, GPU giant Nvidia Corp rolled out its 240-core, heterogeneous Tesla 10-series processors, based on the second generation of the company’s Cuda architecture, which the company said combines the capabilities of many-core processors along with the computational abilities of multi-core processors.

In terms of AMD’s FireStream 9250, the company said it occupies a single PCI slot, consumes less than 150 watts of power, and aims to deliver performance per watt efficiency of up to eight gigaflops per watt.

For more on this topic, see: "Embedded x86: GPU case studies, strengths, shortcomings, and additional perspectives."

Both AMD’s FireStream and Nvidia’s Tesla 10 were designed to run critical workloads such as financial analysis or seismic processing faster than with a CPU alone, which helps them address more complex problems and achieve faster results.

AMD said developers are reporting up to a 55x performance increase on financial analysis codes as compared to processing on the CPU alone, which supports their efforts to make better and faster decisions, and the use of flexible GPU technology rather than custom accelerators assists creators of application-specific systems to enhance and maintain the systems easily.

During a meeting with press last week, Nvidia gave a number of examples that demonstrated the power of its new processor including one from the National Center of Atmospheric Research that created a weather research and forecast model. The center reported a 20% speedup with 1% of the model on Nvidia’s Cuda platform that resulted in a one week savings of analysis time.

The Nvidia Tesla 10-series processor contains 1.4 billion transistors, 1 teraflop of processing power with 240 processing cores. Making up the processing cores is 30 thread processor arrays, each of which contains 8 thread processors, Nvidia explained. The chip also supports full scalar processing with integer and floating point units, as well as single and double IEEE 754 floating point.

Nvidia is rolling out two form factors: a 1U system and a computing processor unit (both pictured below).

Meanwhile, AMD’s FireStream 9250 stream processor includes a second-generation double-precision floating point hardware implementation delivering more than 200 gigaflops, building on its earlier AMD FireStream 9170. AMD said the FireStream 9250’s compact size makes it ideal for small 1U servers as well as most desktop systems, workstations, and larger servers and it features 1GB of GDDR3 memory, enabling developers to handle large, complex problems.

Both companies are offering development kits in support of the new processors.

AMD said it plans to deliver the FireStream 9250 and the supporting SDK in Q3 for $999.

© 2009, Reed Business Information, a division of Reed Elsevier Inc. All Rights Reserved.