At the current stage, it’s preferred to combine CPUs with GPUs, FPGAs or ASIC kernel acceleration cards for heterogeneous computing. In recent years, the conventional way of computation by utilizing CPU frequency and kernel count has encountered a bottleneck in its ability to dissipate heat.

Meanwhile, GPU, FPGA and ASIC cards have increasingly evolved to a higher kernel count and better parallel computing power. As such a system that boasts a high overall ratio of performance / chip area and a performance / watt, the accelerated computation of science, business analysis and engineering application can be improved to a large extent. By combining the servers optimized for sequential processing from GIGABYTE with different types of acceleration cards in collaboration with thousands of small computing cores, heterogeneous computing has obviously become one of the mainstream high performance computing systems today.

To achieve this level of performance, parallel computing across a number of CPUs or GPUs is required. One common type of HPC solutions is a computing cluster, which aggregates the computing power of multiple computers ( referred to as “nodes” ) into a large group. A cluster can deliver much higher performance than a single computer, as individual nodes work together to solve a problem larger than any one computer can easily solve.