← All resources

GPU Programming and Language Design with Chris Lattner

Auto-discovered

Video / channel Advanced

Chris Lattner, Richard Feldman

Visit resource ↗

Description

GPU Programming and Language Design with Chris Lattner is a Software Unscripted podcast episode hosted by Richard Feldman with guest Chris Lattner on Mojo and GPU-oriented language design.

They cover Mojo updates such as collections, pip packaging, traits, and a large open-sourced portable GPU kernel library. Topics include heterogeneous CPU and GPU compilation, sharing numeric code across devices, explicit control versus magic abstractions, tensor cores and memory hierarchy, and how this compares with CUDA and thinner GPU APIs. They also discuss AI-assisted compiler work and specialized hardware after single-thread scaling slowed.

It is for intermediate to advanced systems and language programmers interested in GPU kernels and Python interoperability, not a beginner CUDA lab.

Access is a free YouTube recording of the podcast conversation.

It is useful for a first-principles view of one language spanning host and accelerator code without treating GPUs as a bolted-on dialect.