Skip to content

High-Performance Computing

Introduction

High-Performance Computing (HPC) utilises several fast network-connected servers to perform computationally intensive calculations or process large data sets. Disciplines such as Physics, Medical-Physics, Chemistry, and Engineering benefit from HPC. However, due to the processing requirements of large data sets, such as genome sequencing in fields such as Genetics, Biochemistry, Microbiology, Virology, and Pathology, an increase in HPC utilisation has been observed in the past decade.

Implementing Artificial Intelligence and Machine Learning also demanded fields such as Social Sciences and Humanities to use HPC more frequently. HPC is also often used to automate or streamline scientific workflows or pipelines of data processing.

Using an HPC has several benefits, but the most renowned benefits are saving time, performing several tasks in parallel and processing datasets that were otherwise impossible to process on an ordinary computer or even in a physical laboratory.

Watch the video below to see how an HPC is used in various scientific fields.

These examples are a fraction of the potential use of an HPC and reflect a portion of research projects that utilised the UFS HPC in recent years.

Hardware Spesifications

Compute Nodes

To summarise the components listed in the table below:

  • There are eight compute nodes in the HPC.

  • There are two main different types of nodes. The first four nodes each have eight NVIDIATM RTX 4000 GPUs, and the subsequent four nodes each have four NVIDIATM RTX 5000 GPUs.

  • Each node has two AMDTM EPYC 64 Core CPUs with 256 CPU Core Threads per node.

  • Each node has two terabytes (2 TB) of ECC registered system memory (RAM).

The following table shows the hardware specifications of the compute nodes in the HPC.

Host CPU Model GPUs Total Threads Total RAM
cn0601 2 x AMD EPYC 7773X 1 8 x NVIDIA RTX A4000 256 2 TB
cn0602 2 x AMD EPYC 7773X 1 8 x NVIDIA RTX A4000 256 2 TB
cn0603 2 x AMD EPYC 7773X 1 8 x NVIDIA RTX A4000 256 2 TB
cn0604 2 x AMD EPYC 7773X 1 8 x NVIDIA RTX A4000 256 2 TB
cn0605 2 x AMD EPYC 7773X 1 4 x NVIDIA RTX A5000 256 2 TB
cn0606 2 x AMD EPYC 7773X 1 4 x NVIDIA RTX A5000 256 2 TB
cn0607 2 x AMD EPYC 7773X 1 4 x NVIDIA RTX A5000 256 2 TB
cn0608 2 x AMD EPYC 7773X 1 4 x NVIDIA RTX A5000 256 2 TB
Total (8 Nodes) 16 CPUS 40 GPUs 2 048 Threads 16 TB RAM
Host System Model CPU Model Cores per CPU Threads per CPU GPUs RAM Total Threads Total RAM
cn0601 4124GS-TNR 2 x AMD EPYC 7773X 1 64 128 8 x NVIDIA RTX A4000 32 x DDR4 3200 MHz (64 GB) 256 2 TB
cn0602 4124GS-TNR 2 x AMD EPYC 7773X 1 64 128 8 x NVIDIA RTX A4000 32 x DDR4 3200 MHz (64 GB) 256 2 TB
cn0603 4124GS-TNR 2 x AMD EPYC 7773X 1 64 128 8 x NVIDIA RTX A4000 32 x DDR4 3200 MHz (64 GB) 256 2 TB
cn0604 4124GS-TNR 2 x AMD EPYC 7773X 1 64 128 8 x NVIDIA RTX A4000 32 x DDR4 3200 MHz (64 GB) 256 2 TB
cn0605 4124GS-TNR 2 x AMD EPYC 7773X 1 64 128 4 x NVIDIA RTX A5000 32 x DDR4 3200 MHz (64 GB) 256 2 TB
cn0606 4124GS-TNR 2 x AMD EPYC 7773X 1 64 128 4 x NVIDIA RTX A5000 32 x DDR4 3200 MHz (64 GB) 256 2 TB
cn0607 4124GS-TNR 2 x AMD EPYC 7773X 1 64 128 4 x NVIDIA RTX A5000 32 x DDR4 3200 MHz (64 GB) 256 2 TB
cn0608 4124GS-TNR 2 x AMD EPYC 7773X 1 64 128 4 x NVIDIA RTX A5000 32 x DDR4 3200 MHz (64 GB) 256 2 TB
Total 8 Nodes 16 CPUs 512 Cores 2 x 1 024 Threads 40 GPUs 256 x 64 GB Modules 2 048 Threads 16 TB RAM

Components not contained in the above mentioned table but of some importance:

  • 2 x 800 GB SSD 6 GB/s hard drives per node in RAID 0 configuration.
  • 2 x 3.84 TB NVMe hard drives used for local scratch space.
  • 2 x Mellanox MCX512A-ACAT SFP28 25 GbE Network interfaces.
  • 4 x Redundant 2 000 Watt Titanium Level power supplies per node.

GPU Nodes

Currently, all compute nodes contain GPUs; therefore, no specific GPU nodes are available.

Footnotes:

  1. AMD EPYC 7773X 64 Core with 128 Threads per CPU ↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩↩