nvCOMP Library API examples and benchmarks

Introduction

nvCOMP is a CUDA library that features generic compression interfaces to enable developers to use high-performance GPU compressors and decompressors in their applications. Example benchmarking results and a brief description of each algorithm are available on the nvCOMP Developer Page.

This folder demonstrates the nvCOMP library API usage alongside our performance benchmarks.

More information can be found about the examples here and about the benchmarks here

Supported SM Architectures

Volta (SM 7.0)
Turing (SM 7.5)
Ampere (SM 8.0, SM 8.6)
Ada Lovelace (SM 8.9)
Hopper (SM 9.0)
Blackwell (SM 10.0, SM 10.1, and SM 12.0)

More information can be found about the architectures and compute capabilities on the official NVIDIA website here.

Supported OSes

Linux, Windows

Supported CPU Architectures

x86_64, aarch64

CUDA APIs involved

nvCOMP

Prerequisites

A Linux or Windows system with recent NVIDIA drivers.
CUDA 11.x, CUDA 12.x, or CUDA 13.x toolkit.
CMake 3.18 or above
gcc-8 or above on aarch64 (with C++11 support)
gcc-9 or above on x86-64 (with C++11 support)
The latest nvCOMP release from the NVIDIA developer zone.