NCU Profiler
The NCU Profiler tool lets you open and analyze NVIDIA Nsight Compute (.ncu-rep) reports directly in VS Code. Get a structured view of kernel performance metrics, bottleneck identification, and optimization recommendations—all without leaving your editor.
What is NCU?
NVIDIA Nsight Compute (NCU) is a profiling tool for CUDA kernels. It captures detailed performance data including:- Kernel timing — How long each kernel takes to execute
- Throughput metrics — Compute and memory bandwidth utilization
- Occupancy — How well you’re using the GPU’s compute resources
- Register pressure — Per-thread register usage
- Bottleneck analysis — Roofline model and performance limiters
Features
Report Analysis
Open
.ncu-rep files and browse kernel metrics in a clean, organized interface. Analyze via server or locally.Kernel Summary
See all profiled kernels at a glance with key metrics: duration, throughput, occupancy.
Source View
View CUDA source, PTX, and SASS assembly with line-level performance metrics.
Performance Diagnostics
Get actionable recommendations for optimization based on detected bottlenecks.
Server & Local Analysis
Upload to Wafer’s B200 server or parse locally with your NCU installation.
Create Profiles
Run NCU profiling directly from VS Code when NCU is installed locally.
Requirements
| Requirement | Details |
|---|---|
| Server Analysis | No installation required—reports are parsed on Wafer’s B200 server |
| Local Analysis | NVIDIA Nsight Compute with ncu CLI on PATH |
| For Profiling | NVIDIA Nsight Compute with ncu CLI on PATH |
You can analyze
.ncu-rep reports without NCU installed using server-side analysis. Local analysis and profiling require NCU to be installed on your system.Quick Start
- Select NCU Profiler from the Wafer tool dropdown
- Click Select .ncu-rep file to open an existing report
- Browse the kernel summary and metrics
- Expand the diagnostics panel for optimization recommendations
Analyzing Reports
Learn how to analyze NCU reports in detail →