Reference: https://www.ibm.com/think/topics/hpc


What is it?

Basically throw a bunch of computing resources at a complex problem that would take an eternity to solve. These problems stem from a variety of industries such as chemistry, meteorology, and mathematics. Specific problems including genome sequencing, intensive 3D graphics rendering, automated trading, computational fluid dynamics, and weather predictions.

How does it work?

Parallel Computing
  • Run multiple tasks simultaneously by breaking down the problem into smaller processes
  • HPC clusters may use thousands to millions of processors/processor cores
Computer (HPC) Clusters

"Multiple high-speed computer servers networked with a centralized scheduler1 that manages the parallel computing workload"

  • Computers, aliased as 'nodes', more prevalently use GPU's but may use multi-core CPU's to perform mathematical calculations, ML2 models, and other intensive tasks
Other HPC Components
  • Networking
  • Memory
  • Storage and File Systems
  • Low Latency
Message Passing Interface (MPI)
  • Standard library and protocol for parallel computer programming enabling users to communicate between nodes

HPC vs. Quantum Computing

  • Conventional bits (0 and 1) rather than qubits (quantum states existing as superpositions of 0 and 1)
  • Quantum Computing would be a more efficient way of solving complex problems (e.g. molecular behavior)
  • Quantum Computing is not expected to replace HPC, rather work in harmony to achieve optimal performance

Why do I want to study it?

Consistently, I had wanted to understand computers from a fundamental level. The best way to learn about computers is to start from the bottom up. From transistors to low level programming. After taking my Computer Organization class at VT3 and designing a payload for my rocket design team, I wanted to pursue programming closer to the hardware rather than the software. Although I enjoy building things for the community, that is more of a hobby rather than a specialized skill.

Hopefully, I will build some interesting projects in C++ using CUDA and the OpenMP library to learn about parallelism.


Footnotes

  1. a software component that decides which computer server performs which computing task, and when

  2. Machine Learning

  3. Virginia Tech