API Comparison

nvCOMP offers four API tiers. Each targets a different integration depth and/or programming language.

Which API Should I Use?

API	Language	Complexity	Memory Management	Choose When
Python Codec	Python	Low	Automatic	You work in Python and want the simplest path to GPU compression.
C++ Manager (HLIF)	C++	Low–Medium	Automatic (Manager handles temp buffers, chunking if desired)	You need C++ integration with managed state.
C Batched (LLIF)	C/C++	High	Manual (user allocates temp, output, and status arrays)	You need maximum control over GPU resources, or you are integrating into a C codebase. You manage chunking, temp workspace, and device pointer arrays yourself.
nvCOMPDx (Device Extensions)	CUDA C++	High	Manual (compile-time descriptors, user-managed shared memory)	You want to call (de)compression from inside your own CUDA kernel at warp or block granularity, avoiding separate kernel launches. Only for ANS and LZ4.

For quick-start examples of each API, see:

• High-level C++ Quick Start Guide (HLIF)

• Low-level C Quick Start Guide (LLIF)

• Python API (Python)

• nvCOMPDx Introduction