Skip to main content

​
Benchmarking

Test your CUDA kernel performance and compare different implementations.

​
Quick Start

1

Open Benchmark Panel

Click “Benchmark” in the bottom panel or use command palette: “CUDA: Open Benchmark View”.
2

Configure Benchmark

Set iterations, warmup runs, and data sizes in the configuration view.
3

Run Benchmark

Click “Run Benchmark” button and watch results appear in real-time.
4

Compare Kernels

Select two benchmarked kernels to see side-by-side performance comparison.

​
How to Use

Benchmark Configuration

​
Configuration View

Set your benchmark parameters:
  • Iterations: Number of times to run (e.g., 100)
  • Warmup Runs: Initial runs to stabilize GPU (e.g., 10)
  • Data Sizes: Small, medium, large test configurations

​
Running Benchmark

  1. Configure your settings
  2. Click “Run Benchmark” button
  3. Watch progress bar as results stream in
  4. See live metrics updating

​
During Execution

Benchmark Running
  • Progress indicator shows completion percentage
  • Results appear in real-time
  • Current iteration displayed
  • Stop button available to cancel

​
What You See

​
Live Results

  • Running: Shows current iteration (e.g., “Running 45/100”)
  • Time Graph: Real-time performance visualization
  • Statistics: Live mean, min, max updates

​
Final Results

  • Execution Time: Average kernel runtime
  • Statistical Analysis: Mean, median, standard deviation
  • Performance Metrics: Throughput, efficiency
  • Export Button: Save results

​
Comparing Kernels

Benchmark Comparison
  1. Benchmark your original kernel
  2. Benchmark your optimized version
  3. Click “Compare” button
  4. View comparison:
    • Side-by-side metrics
    • Speedup percentage (e.g., “2.3x faster”)
    • Winner highlighted in green

​
Tips

Best Practices
  • Use at least 100 iterations for accuracy
  • Always include warmup runs
  • Keep data sizes consistent for comparison
  • Benchmark before and after optimization
Important
  • Close other GPU applications for accurate results
  • Let GPU stabilize between tests
  • Use same configuration when comparing kernels

​
Screenshots Needed

  1. Benchmark configuration view - Shows settings panel
  2. Benchmark running - Progress bar and live results
  3. Kernel comparison - Side-by-side performance metrics

​
Next Steps