AMD has announced its latest FirePro accelerator board, boasting 32GB of RAM and up to 5.24 teraFLOPS of single-precision compute from a single GPU.
Designed for the high-end server and high-performance computing (HPC) markets, the FirePro S9170 packs 32GB of GDDR5 memory on a 512-bit bus with 320GB/s bandwidth - the first server GPU to feature such a hefty chunk of VRAM, and a big reason behind AMD's claim that the new card can beat Nvidia's competing equivalent - the Tesla K40 - by 40 per cent in double-precision and DGEMM heavy workloads.
The company's claimed 5.24 teraFLOPS single-precision and 2.62 teraFLOPS double-precision peak compute performance is undeniably impressive: the Nvidia Tesla K40 offers just 1.43 teraFLOPS in double-precision mode, and even the dual-GPU Tesla K80 hits a high of 1.87 teraFLOPS. The performance comes at a cost of power, of course: despite having only a single GPU, the FirePro S9170 demands 275W - far higher than the 235W thermal design profile (TDP) of the Tesla K40, and only just below the 300W of the dual-GPU K80.
'Today the best GPU for compute just got better with the introduction of the AMD FirePro S9170 server GPU to complement AMD's impressive array of server graphics offerings for high performance compute environments,' crowed Sean Burke, corporate vice president and general manager of AMD's Professional Graphics group, at the unveiling. 'The AMD FirePro S9170 server GPU can accelerate complex workloads in scientific computing, data analytics, or seismic processing, wielding an industry-leading 32GB of memory. We designed the new offering for supercomputers to achieve massive compute performance while maximizing available power budgets.'
Gigabyte and Asus have already come forward as designing systems around the new card, with the former designing a 2U rack-mount server with up to eight S9170s for a total of 256GB of GDDR5 VRAM accessible to the system. 'This unprecedented density of GDDR5 memory will help our customers offer very attractive machines to industries processing large datasets,' Gigabyte's Alex Liu claimed.
Pricing for the card, which is to hit the market in the third quarter of this year, has yet to be confirmed - but expect it to be very much at the top of the tree.