This repository contains the HMM demos of the Simplifying GPU Application Development with HMM (Heterogeneous Memory Management) blogpost.
The HMM requirements are described in the CUDA 12.2 Release Notes.
The demos require a system with ATS or HMM enabled; this can be verifyied by querying the Addressing Mode using nvidia-smi:
$ nvidia-smi -q | grep Addressing
Addressing Mode : HMMThe demos are available in the src/ directory. On systems with docker installed, they can be run as follows:
./ci/runSee LICENSE.