Supermicro GPU System to Enhance Research Abilities at SZTAKI, the Institute for Computer Science and Control

Based in Budapest, the Institute for Computer Science and Control is a research organization that focuses on a range of research domains, including computer science, engineering, information technology, and intelligent systems. Advanced GPU capabilities are required for many research areas, including medical image processing, machine perception for autonomous vehicles and robotics, and natural language modelling.

The algorithms that we use demand a scalable computing environment for CPU and GPUs. Some of the algorithms required more than one GPU to complete the task in a reasonable amount of time.

The team at SZTAKI investigated several servers from various system manufacturers and settled on using the Supermicro A+ 4124GO-NART GPU server. This system contains two AMD EPYC™ CPUs and eight NVIDIA® HGX A100 GPUs, and 1 TB DRAM. Using multiple GPUs that are connected together with NVIDIA® NVLink® and NVSwitch™, the GPUs can communicate at up to two terabytes/second. AMD EPYC™ CPUs, operating at 3.2 GHz, run the applications, interacting with the GPU sub-system as needed. The NVIDIA® HGX™ A100 8-GPU 40GB contains the latest Tensor Core GPUs, enable applications to run faster and return results to the researchers faster. The teams at SZTAKI have been able to increase their pace of research due to the new Supermicro system.

In this session, we demonstrate how the Supermicro 4124GO-NART GPU server drastically increased the performance of specific medical image processing tasks, including the segmentation of cardiac MRI, dental CT and chest CT data enrichment by ProGAN, and other 3D convolutional network applications.