AMD is upping the ante in GPU compute with its new FirePro S9170 server GPU that features 32GB of on-board GDDR5 memory. This is double that of its predecessor, the FirePro S9150, and 8GB more than Nvidia’s top end GPU compute board the Tesla K80.
According to AMD, 32GB of memory, coupled with 320GB/s memory bandwidth will enable firms to process larger and more computationally complex workflows than before. This has sometimes been a barrier for using GPUs in High Performance Computing (HPC).
“There are some HPC workloads which require as much data as possible to stay resident on the device, and so the 32GB of memory provided by AMD FirePro S9170, the largest available on a single GPU, will enable the acceleration of scientific calculations that were previously impossible,” said Simon McIntosh-Smith, head of the Microelectronics Research Group at the University of Bristol.
The FirePro S9170 is rated at 275W, but AMD has said that there will also be a 235W model for compatibility with a wider range of servers.
Elsewhere, the specs are quite similar to the AMD FirePro S9150, which will continue to be sold by AMD. There is a slight increase in compute performance – single precision (5.24 TFLOPS vs 5.07 TFLOPS) and double precision (2.62 TFLOPS vs 2.53 TFLOPS).
For product development, AMD’s FirePro server cards are best suited to GPU clusters used for ray trace rendering or Computer Aided Engineering (Finite Element Analysis and Computational Fluid Dynamics). The software must be compatible with OpenCL.