Nsight Copilot
Nsight Copilot for VS Code is NVIDIA's accelerated computing programming assistant. It provides intelligent code suggestions and assistance for CUDA development.
Overview
Nsight Copilot for VS Code is NVIDIA’s accelerated computing programming assistant. It provides intelligent code suggestions and assistance for CUDA development.
We are thrilled to announce the General Availability (GA) of the Nsight Copilot VS Code extension, along with new models. Combined with our specialized CUDA-aware LLM inference models, it aims to deliver the best coding experience for CUDA developers. Model performance is benchmarked using our ComputeEval framework to assess CUDA-related task proficiency.
Key Features
- Public General Availability (GA): Nsight Copilot 2025.1 is now public! Any developer with a free NVIDIA Developer account can install the extension from the VS Code Marketplace. Your feedback during the private preview was instrumental in reaching this milestone.
- Upgraded Chat Model: The chat window has been upgraded with the powerful
gpt-oss-120b model to provide significantly higher-quality answers for your CUDA queries.
- AI Quality backed by ComputeEval benchmark for evaluating CUDA awareness.
- New CUDA-Aware Autocomplete Model: A new autocomplete model,
nvidia/CUDA-autocomplete, is now available. The Autocomplete Model is fine-tuned on CUDA to provide highly relevant code completions. To use the Autocomplete Model feature follow the steps mentioned below.
Requirements
- NVIDIA Developer Account. You can sign up for an NVIDIA account here.
Quick Start
- Locate the Nsight Copilot VS Code Extension in the Visual Studio Code Marketplace and install it (if you have previously installed it, please make sure the auto update of the extension is on, and it’ll automatically update to the newest version).
- Log into Nsight Copilot extension with your NVIDIA developer account (How to sign up for an NVIDIA Developer account).
Explore Nsight Copilot in action
Put your new CUDA-aware assistant to the test. For example, try asking:
- "Please help me understand core CUDA concepts."
- "How do I perform summation of an array using the CUB library?”
To sign up for an NVIDIA Developer account
- Start here to create an NVIDIA Developer account email.
How to Use the New Autocomplete Model
- In the bottom status bar, click on the "Nsight Copilot" text.
- Select nvidia/CUDA-autocomplete from the pop-up list.
- Important for Cursor users: Cursor has its own AI autocomplete that will override this setting. To use the nvidia/CUDA-autocomplete model, you must disable the Cursor’s native autocomplete feature first: hover over Cursor Tab in the bottom status bar, and click “Disable globally” or “Disable for cuda-cpp”.
Support
- Your feedback is vital to shaping the future of AI in accelerated computing!
- What works well, what doesn't, and any unexpected behavior?
- Noticed anything wrong with the suggested code? If possible, please include the prompt and generated code snippet.
- Features you’d like to see.
- Please share your feedback at Nsight-Copilot-Feedback@nvidia.com
- Reach out to us for feedback and questions via our developer forum.
Data and telemetry
This extension collects all prompts and output, and additional usage data as described in the NVIDIA Software License Agreement to improve NVIDIA products and services, including AI models. Do not upload any confidential information or personal data. Your use is logged for security, fraud or abuse monitoring and shared with third party service providers for that purpose.
This extension respects the "telemetry.enableTelemetry" setting, for more info see Visual Studio Code Telemetry.
License
This project is used under NVIDIA Software License Agreement.
Third-Party Licenses
This project uses various third-party libraries. For a complete list of licenses, please see THIRD_PARTY_LICENSES.txt.
