AMD has revealed the world’s first 32 GB server GPU with support for OpenCL 2.0. Based on the company’s second generation Graphics Core Next technology, the FirePro S9170 is designed for DGEMM heavy double-precision workloads, and is the latest addition to AMD’s FirePro server GPU lineup.
FirePro S9170 is capable of up to 5.24 TFLOPS of peak single precision compute performance, and up to 2.62 TFLOPS of peak double precision performance. Single precision just slightly edges out Nvidia’s best single GPU, but double precision outperforms the Tesla K40 by nearly a full TFLOP, and even nips at the heels of Nvidia’s Tesla K80 dual GPU flagship, which can manage 2.91 TFLOPS at the highest boost clock.
FirePro S9170’s 32 GB of memory has been paired with a 512-bit memory bus and delivers 320 GB/s of memory bandwidth. AMD said that the extra memory over the competing GPU helps to improve overall workload speed and system responsiveness while working with large datasets.
AMD is positioning the new FirePro card as an ideal solution for the compute-intensive workflows that are found in the oil and gas industries, academic and government bodies, and deep neural network compute cluster development. The company said scientific computing, data analytics and seismic processing can all benefit from the added memory on the FirePro S9170.
AMD’s FirePro S9170 server GPUs support the latest version of OpenCL and are ready to use with OpenMP and OpenACC developer tools, which are planned for Q3 availability.
