We have a high-end platform, focused on housing up to 8 NVDIA Tesla K series cards, the most powerful focused on scientific calculation.
This node, designed to be rack mounted, offers great density, since it only occupies 3U and can function as a single machine or be a node in a cluster.
GPU-based HPC systems are becoming increasingly popular because they bring the power of an entire cluster to a single machine.
In fact, this solution can offer around 70 Tflops in single precision and 23 Tflops in double precision, thanks to the new Tesla K80. These powers, until recently, were only possible in clusters that took up a full rack and cost 5 times as much. In addition, logically when we talk about cluster solutions, we must increase other costs such as communications, rack, cooling, etc.
In addition, the fact that all the calculations are handled on a single machine also reduces issues of management complexity and latency between nodes. Applications like Amber14, Gromacs, NAMD, LSMS, Chroma, River flow 2D, all of them in the scientific computing environment, improve their real performance between 5 and 15 times running on GPU vs. CPU.
