Wafer logo
Back to all blogs
January 13, 2026
3 min read
Wafer Team

Introducing ROCprofiler Compute: AMD GPU Profiling in Your IDE

Profile AMD GPUs directly in VS Code and Cursor. View hardware metrics, roofline analysis, and kernel stats — all without leaving your editor.

AMDROCmProfilingMI300XGPU
ROCprofiler Compute Architecture View

TLDR: ROCprofiler Compute brings AMD GPU profiling into your IDE. View hardware metrics, roofline analysis, and kernel performance data directly in VS Code or Cursor.

Today we're launching ROCprofiler Compute support in Wafer.

If you're profiling AMD GPUs (MI300X, MI250, etc.), you can now view your rocprof-compute results directly in your editor with an interactive GUI.

What's included

1. GPU Architecture Overview

See your kernel's performance at a glance. The architecture diagram shows every hardware block — instruction dispatch, compute units, caches, memory bandwidth — with real metrics from your profiling run.

GPU Architecture Diagram

2. Roofline Analysis

Understand where your kernel sits relative to hardware limits. The roofline plot shows arithmetic intensity vs. performance for L1, L2, HBM, and peak compute.

Roofline Analysis

3. Kernel Statistics

View top kernels by execution time, dispatch lists, and per-kernel breakdowns. Sort and filter to find the hotspots.

Top Stats View

4. System Info

Full hardware details: GPU model, compute units, cache sizes, clock speeds, ROCm version — everything you need to understand your target hardware.

System Info

How to use it

Option 1: View existing profiling results

If you already have rocprof-compute results:

1. Open Wafer in VS Code/Cursor

2. Select "ROCprofiler Compute" from the tools menu

3. Browse to your workload folder (must contain sysinfo.csv)

4. Click "Launch GUI"

Option 2: Profile and analyze with wafer-cli

Profile your application:

bash
wafer rocprof-compute profile --name my_kernel -- './my_hip_app'

Analyze results:

bash
wafer rocprof-compute analyze workloads/my_kernel

Launch GUI:

bash
wafer rocprof-compute analyze workloads/my_kernel --gui

Requirements

  • ROCm 7.0+ with rocprof-compute 3.2+
  • AMD GPU with ROCm support (gfx908, gfx90a, gfx940, gfx941, gfx942, gfx950)

The GUI viewer works on any platform — it only reads CSV files from your profiling run. You can profile on a remote AMD machine and view results locally.

Try it out

Install today and let us know what you think.