Groq: Traditional HPC vs converged HPC (Booth 119)
Performance and complexity comparison between purely mathematical algorithmic calculations compared to AI/ML model-based calculations, using JAX-CFD, a CFD solver written in JAX. The program solves incompressible flows on structured grids with explicit time integration.
We also show the complexity of creating a DNS-based algorithm vs an ML-based one, and the ease of model conversion to The Groq Chip - often with a single line of code!
We will demonstrate CPU, GPU (A100), and Groq TSP comparisons for Performance and accuracy of each demo with live visual output.
- Chip with massive performance and acceleration compared to GPUs
- Groq's unique and extremely rapid and simple programming environment is demonstrated live.
- See how AI in HPC is a viable and better replacement for some compute-intensive applications.
- Demonstrate how versatile a chip can be from pure algorithmic HPC to converge HPC with ML.