HIP Memory Manager (hipMM) documentation

HIP Memory Manager (hipMM) documentation#

2026-07-29

1 min read time

Applies to Linux

The HIP Memory Manager (hipMM) provides advanced GPU memory management utilities used across the AMD Data Science libraries. Based on the RAPIDS® Memory Manager (RMM 25.10), hipMM supports efficient allocation, pooling, and data movement, ensuring stable performance in complex, multi-library GPU workflows.

hipMM 4.0.0 improves memory usage efficiency for workloads that leverage hipDF, hipGRAPH, hipRAFT, and hipVS.

The hipMM code is open and hosted at AMD-Ecosystem/hipMM.

The hipMM documentation is structured as follows:

To contribute to the documentation refer to Contributing to AMD Data Science.

You can find licensing information on the Licenses page.