A new press release from AMD announces their new AMD FireStream 9350 and 9370 GPU compute accelerators, apparently feeling the hurt from the new NVidia Fermi offerings as the entire press release focuses on the 2x improvement in Single and Double Precision performance.
“Our customers’ face increasingly demanding enterprise data center requirements for power consumption, cooling, and floor space, and at Supermicro our goal is to deliver solutions that offer maximum performance per watt, per square foot and per dollar,” said Don Clegg, vice president of Marketing at Supermicro. “By offering GPU compute accelerators in combination with powerful multi-core CPUs, we are keeping pace with our customers’ demands. The AMD FireStream 9350 and 9370 compute accelerators are a natural fit for our industry-leading server solutions.”
They claim that the new 9370 offers 2.64 TeraFlops of single precision performance, and 528 GigaFlops of double-precision performance, all with 4GB of GDDR5 memory and 225Watts of power. That puts a 5x boost on single-precision over double-precision, not too shabby. These numbers still beat the max figures of the NVidia cards, but the big question is “What possibly can achieve that maximum throughput in a useful way?” The short answer is: nothing.
It definitely looks like a great card from ATI, but I’m sure in practical benchmarks between it and the new Fermi-driven NVidia offerings, the NVidia will probably still edge it out due to CUDA support and better scheduling and kernel optimizations.
