liveBook · Manning

ROCm

As explained in GPU Programming with Triton

ROCm is AMD's open-source GPU computing stack comprising driver, runtime, and compiler-toolchain components.

ROCm is AMD’s open-source GPU computing stack. For Triton, it supplies the runtime needed to execute kernels on AMD hardware; installing the Python package alone is insufficient. section 1.1.1

Triton compiles each kernel for the active backend, producing AMDGPU IR on AMD systems. Without the matching ROCm runtime, compilation fails even when Triton itself is installed. section 1.1.1

Running a Triton kernel on AMD

A Python function decorated with @triton.jit can remain in the Python workflow while targeting AMD hardware, provided ROCm is available on the machine. section 1.1.1

How GPU Programming with Triton covers it

How deep it goes
Identifies the runtime requirement and AMDGPU IR target, but does not detail ROCm’s internal components.
Examples
Uses @triton.jit functions, AMD hardware, ROCm availability, and AMDGPU IR as concrete references.
What it assumes
Assumes Python and Triton usage while establishing that the corresponding hardware runtime must also be installed.
What you can do after
Operate Triton kernels on AMD hardware by installing and providing the matching ROCm runtime.
How it connects
Connects Triton, GPU kernels, Python DSLs, CUDA, and AMDGPU IR through backend-specific compilation.

This explainer was generated by Manning's knowledge graph from the text of GPU Programming with Triton. The section links point to the parts of the book it draws on.