Vendor-agnostic Accelerated Compute Glossary - Task 052

Fit to programme

This task has been identified by the working groups as part of the agenda behind WP 2.3.

The task number is 052.

Summary

Accelerated programming is full of jargon: warps, SMs, CUDA cores, and so on. Every programming model (OpenMP, CUDA, SYCL) adds its own terminology on top of this already crowded landscape.

When teaching programming for accelerated compute, it would help to have a table that explains each concept individually and relates it to others. For example, a learner should be able to look up “SIMT” and, after a brief explanation, see how it connects to SIMD in Flynn’s taxonomy.

These connections are frequently missing from other tutorials, which tend to present terminology from a single vendor’s point of view rather than grounding it in general computer science concepts.

Ideally, the material would be presented as an interactive map, similar to https://shareing-dri.github.io/hpc-concepts-course/. Similar material already exists (e.g., https://modal.com/gpu-glossary), but it doesn’t yet link back to the underlying computer science fundamentals.

Outcomes

Training material