Pages

November 08, 2007

AMD FireStream 9170 is double precision CPU/GPU hybrid with 500 Gigaflops

AMD's FireStream 9170 chipset includes 660 million transistors and 320 processing units and will be made by chipmaker TSMC. The FireStream 9170 is a step on the way to AMD's Fusion project, which the company says will combine a graphics processor and general processor on the same piece of silicon. AMD hopes to release Fusion in 2009. It is similar to the Nvidia Tesla, which is currently single precision by will likely have a double precision version in 2008.

The 500 gigaflops is about 100 times the performance of one of its dual core Opterons.

When it becomes available during the first quarter of 2008, the AMD FireStream 9170 will be the world's first stream GPU with double-precision floating point technology created specifically for scientific and engineering calculations, the chipmaker said.

A single card solution, the 9170 offers 2 GB of memory and as much as 500 gigaflops of computing power. That, according to AMD, rivals the power of many supercomputers today and also provides dramatic acceleration for critical algorithms.

Unlike AMD's other processors, including the Opteron, the FireStream 9170 is built with a 55 nanometer process technology and uses less than 150 watts of power. Consuming less power equals reduced heat dissipation, which allows the 9170 to function in dense design configurations.


the chip is expensive, priced at $1,999, it boasts 500 gigaflops of computing power for critical algorithms.

Together with the FireStream product launch is the release of its SDK. This software development kit introduces the Compute Abstraction Layer API, which builds on the Close to the Metal (CTM) driver interface introduced in 2006. For high-level development, AMD is announcing Brook+, a tool providing C extensions for stream computing based on the Brook project from Stanford University. In addition, AMD also plans to support the AMD Core Math Library (ACML) to provide GPU-accelerated math functions, and the COBRA video library accelerates video transcode.

FURTHER READING
A blogger at Intel Research, Anwar Ghuloum, attacks the designs of GPGPUs and claims that the performance can only be used for some applications. A lot of science applications (molecular modelling) can use it so that is plenty important for driving the pace of technological progress.

2 comments:

ChemSpiderMan said...

There is a discussion presently going on regarding a comparison of FPGA, GPU and the Cell/B.E processors for the purpose of computational computing. See here: http://www.simbiosys.ca/blog/2008/05/03/the-fast-and-the-furious-compare-cellbe-gpu-and-fpga/

I am interested in your opinions on GPU versus Cell processor.

bw said...

Thanks for the link. It has a good analysis.

I think the cell processor and the GPGPUs and FPGA acceleration are all very promising.

IBM provides workstation support for the Cell processor or you can cobble together PS3s as others have done.

The specific application will determine which is superior. The article you provided talked about a 50 times speed improvement from the cell. Some GPGPU and FPGA situations have several hundred times acceleration.

All three methods will be used for new supercomputers.

I will try to examine each and the issues for the three more closely.