Nsight Visual Studio Code Edition installs statistics

Nsight Visual Studio Code Edition

Overview  |  Documentation  |  Forum  |  Code Samples

Nsight Visual Studio Code edition is an extension for Visual Studio Code that provides support for CUDA development, including features such as Intellisense, debugging, debugger views, and productivity enhancements.

The extension is available on the Visual Studio Marketplace and on the Open VSX Registry for VSCodium and other compatible editors. Linux x86_64 and Linux arm64 hosts are supported.

Benefits

• Higher Productivity<br> Using CUDA-aware code completion improves the overall development experience and enables users to save time and effort when writing code.

• Interactivity<br> Debug CPU and GPU code in the same session using familiar editor controls and debugger views. Investigate CUDA application behavior alongside your source code without switching to a separate debugger interface.

• Remote Development Support<br> Nsight Visual Studio Code Edition enables developers to implement CUDA code in various cluster environments such as Virtual Machines or remote Docker containers. It also supports code development for Linux systems via the Remote – WSL extension. For supported remote debugging configurations, Nsight VSCE can automatically start cuda-gdbserver.

• Free<br> As with other Nsight tools from NVIDIA, Nsight Visual Studio Code Edition is offered free of charge. We love it when your code works better and is delivered sooner. Enjoy!

Key Features

• CUDA C++ Language Support<br> Edit CUDA C++ with syntax highlighting and CUDA-aware code completion, navigation, references, and rename support using either Microsoft's C/C++ extension or clangd.

• CUDA Kernel Breakpoint Support and Kernel Execution Control<br> Break into a debugging session in CPU or GPU device code using standard breakpoints, including support for conditional breakpoints with expression evaluation. Breakpoints can also be set on individual instructions in the Disassembly view. GUI controls allow you to step over, into, or out of statements in the source code, just like normal CPU debugging. Breakpoints are evaluated for every kernel thread and will halt execution when any thread encounters them.

• GPU and Kernel State Inspection<br> When your application is paused in the debugger, inspect call stacks, application variables, CUDA built-in variables, and registers, including PTX special registers. Use the Watch view to track expressions and focus controls to select the CUDA thread being inspected.

Want to know more?

• See the Nsight VSCode Edition spotlight video<br> This Nsight VSCode Edition spotlight shows you how Nsight VSCode Edition fits in with the other NVIDIA IDE debuggers and can be set up in Microsoft's Visual Studio Code. Then, you'll see all the key features in action. You're going to love it!

• See Nsight VSCode Edition demonstrated at GTC'21<br> GTC'21 Video On Demand: Latest Enhancements to CUDA Debugger IDEs

• View the Microsoft announcement for the VSCode extension by Nsight VSCE<br> CUDA Support in Visual Studio Code with Julia Reid

• Read the blog posting<br> Announcing NVIDIA Nsight Visual Studio Code Edition: New Addition to the Nsight Developer Tools Suite

• Visit the Nsight VSCode Edition overview page<br> The Nsight VSCode Edition overview page is your information hub for general information, availability, videos, and other links to other NVIDIA tools for GPU code development.

Requirements

• Visual Studio Code 1.101 or later (or a compatible editor such as VSCodium).

• CUDA Toolkit: Install the CUDA Toolkit (13.3 recommended) to get important tools for CUDA application development including the NVCC compiler driver and cuda-gdb, the NVIDIA tool for debugging CUDA.

• Optional CUDA C++ language provider: for code intelligence, install either clangd or Microsoft's C/C++ extension. Neither is a hard dependency; Nsight guides the setup for whichever is available.

Quick Start

Open or create a new CUDA application. <br> Configure CUDA C++ editing support<br> Open a workspace that contains .cu or .cuh files. Nsight can automatically prompt to configure CUDA C++ language support, or you can run CUDA: Configure CUDA C++ Language Support from the Command Palette.

For clangd, Nsight can generate a project-owned .clangd file and will not overwrite an existing one. If a .clangd file already exists, run CUDA: Generate .clangd and confirm the prompt to create a .clangd.nsight.example file for comparison (only created if one does not already exist).…