CUDA Ada Compatibility Guide

Type: Guide Tags: NVIDIA, CUDA, Ada, compatibility, PTX, cubin, NVCC, compute capability Related: CUDA-Compatibility, CUDA-Ada-Tuning-Guide, NVIDIA-Ada-Lovelace-Architecture, NVIDIA-Ampere-Architecture, CUDA-Programming-Guide, NVCC, PTX-ISA Sources: https://docs.nvidia.com/cuda/ada-compatibility-guide/index.html Last Updated: 2026-04-29

Summary

The CUDA Ada Compatibility Guide is NVIDIA’s application note for running CUDA applications on Ada-generation GPUs. It covers PTX/cubin compatibility, the relationship between Ampere and Ada binaries, PTX JIT verification, and architecture build targets.

Detail

Purpose

Ada is closely related to Ampere for CUDA compute compatibility but adds new instructions and architecture targets. The compatibility guide helps developers decide when existing Ampere-targeted binaries can run on Ada and when they should rebuild with Ada-native cubins.

Key guidance

  • CUDA binaries may contain native cubin code, forward-compatible PTX, or both.
  • Ampere binaries can run on Ada in documented cases, but Ada-native binaries do not run backward on Ampere.
  • CUDA_FORCE_PTX_JIT=1 can verify whether an application has usable embedded PTX.
  • Explicit -gencode targets matter when supporting Volta, Turing, Ampere, Ada, and later GPUs.

NVIDIA context

Use this page for Ada migration and build compatibility. Use CUDA-Ada-Tuning-Guide for performance tuning after compatibility is established.

Connections

Source Excerpts

  • NVIDIA’s Ada compatibility guide describes PTX/cubin behavior, Ampere-to-Ada compatibility, and -gencode build targets.

Resources