CUDA C++

• Get hardware architecture

# x86_64: 64-bit Intel/AMD
# aarch64/arm64: 64-bit ARM (Apple Silicon, Graviton, many ARM servers)
# armv7l: 32-bit ARM
# i686: 32-bit x86
uname -m

• Get Ubuntu version

cat /etc/os-release

• download CUDA toolkit

deb (network) preferred

Installs CUDA via NVIDIA’s apt repo; integrates cleanly with Ubuntu and supports normal apt upgrade/remove. Best default for reproducible setups.

deb (local)

Downloads a local CUDA repo snapshot and installs via apt; useful for offline or restricted environments.

runfile (local)

Standalone installer that bypasses apt and installs CUDA directly; more control, doesn’t integrate with your package manager. Harder to uninstall cleanly. Can conflict with distro drivers.

C++ installation

# on ubuntu
sudo apt update && sudo apt install -y build-essential

Check & Specify C++ standard

• 201103L → C++11
• 201402L → C++14
• 201703L → C++17
• 202002L → C++20
• 202302L → C++23

// To see which standard macro is active, do this:
#include <iostream>

int main() {
    std::cout << __cplusplus << "\n";
}

• To specify the C++ standard:

# use std flag

g++ -std=c++20 test.cpp -o test