Silicon Mechanics, a Seagate Cloud Builder Alliance partner, recently announced that is offering the NVIDIA Tesla K40 GPU accelerator in its GPU Solutions products. The K40 offers significant memory advantages over the K20(X), making it ideal for scientific research and demanding, high-performance computing workloads
Silicon Mechanics products such as the 2U Rackform iServ R354.v4 and the 4U Hyperform R2504.v4 can support up to four Tesla K40 GPUs. The new accelerator significantly boosts application speeds by using a larger memory pool than its predecessors. It is 40 percent faster than the K20(X) and can convert power headroom into a performance acceleration. Like other Tesla GPUs, the K40 is built upon the NVIDIA Kepler architecture.
Clocked at 745 MHz, the K40 has 12GB of VRAM, double the amount in the K20(X). In a post for AnandTech, Ryan Smith stated that one of the most significant changes in the K40 is the implementation of 4Gbit GDDR5 modules. These only recently entered mass production and should facilitate notable improvements in capacity.
“NVIDIA has a number of customers in the compute market who are bottlenecked at either the algorithmic or data set levels by memory capacity,” explained Smith. “So the additional capacity should offer a welcome improvement for those users, and unlock at least a few more workloads that couldn’t properly fit inside of 6GB.”
With Silicon Mechanics’ NVIDIA-enabled solutions, enterprises can better address a wide range of challenging workflows. As more companies roll out data analytics and business intelligence initiatives, they will need high-performing hardware that can support information gathering and collation. In addition, a K40 GPU is a valuable asset for advanced tasks in video and signal processing, computational physics, biology, chemistry and finance.