How to Benchmark OpenCL GPU Acceleration in GIMP

This article outlines how to configure, test, and benchmark OpenCL hardware acceleration across multiple compatible GPU devices inside GIMP. You will learn how to enable OpenCL within the interface, target specific graphics adapters on multi-GPU systems, utilize repeatable GEGL-based rendering workloads, and monitor real-time processing metrics using GIMP’s internal diagnostic tools.

1. Enable OpenCL in GIMP

Before running performance tests, ensure OpenCL support is active inside the application:

  1. Launch GIMP and open Edit > Preferences (or GIMP > Preferences on macOS).
  2. In the left sidebar, select System Resources.
  3. Under the Hardware Acceleration section, check the box labeled Use accelerator (OpenCL).
  4. Click OK and restart GIMP to apply the changes.

If the option is grayed out, ensure that proper OpenCL runtime drivers (such as ROCm for AMD, CUDA Toolkit for NVIDIA, or the Intel OpenCL compute runtime) are installed on your operating system.

2. Configure the Real-Time Dashboard

GIMP includes a built-in telemetry panel to observe processing throughput during computationally intensive tasks:

  1. Navigate to Windows > Dockable Dialogs > Dashboard.
  2. Click the menu icon (a small arrow pointing right) in the top-right corner of the Dashboard dock.
  3. Select Add Metric and enable CPU, Memory, and Swap.
  4. Note the activity line during filter execution to verify when processing begins, sustains, and concludes.

3. Switch Between Multiple GPU Devices

GIMP relies on the GEGL (Generic Graphics Library) engine for OpenCL computation, which automatically initializes the default OpenCL platform and device detected by your system. To benchmark separate GPUs individually (for example, an integrated GPU versus a dedicated card, or two distinct dedicated cards), use system environment variables prior to starting GIMP.

Linux

Identify available OpenCL platforms and devices using the terminal:

clinfo -l

Launch GIMP targeted to a specific platform or device index:

# Example: Select platform 0 or dedicated GPU instance
CUDA_VISIBLE_DEVICES=0 gimp
# Or for AMD:
DRI_PRIME=1 gimp

Windows

You can force Windows to assign a specific GPU to GIMP via the OS settings:

  1. Open Windows Settings > System > Display > Graphics.
  2. Add the gimp-2.10.exe (or current version) executable.
  3. Click Options and set the GPU preference to either Power saving (integrated GPU) or High performance (discrete GPU).

4. Run a Standardized Benchmark Workload

Because GIMP does not possess an automated synthetic benchmark button, use a controlled, repeatable GEGL filter pipeline to measure execution duration across hardware setups.

  1. Prepare the Test File: Create an uncompressed canvas at a high resolution (e.g., 3840×2160 or 7680×4320) filled with continuous noise via Filters > Noise > Plasma. A high-pixel-count image ensures compute kernels remain saturated long enough to produce measurable data.
  2. Select an OpenCL-Accelerated Filter: GEGL operations benefit most from OpenCL. Go to Tools > GEGL Operation... and select an intensive algorithm such as gegl:gaussian-blur or gegl:motion-blur. Alternatively, use Filters > Blur > Gaussian Blur... directly.
  3. Set Static Parameters: Configure standard values (e.g., a radius of 100.00 px).
  4. Execute and Measure:

    • Apply the filter while recording the duration from execution start to completion.

    • For precise timings, launch GIMP from the terminal or command line with the verbose performance flag:

      gimp --verbose

    • Terminal logs will print detailed GEGL execution and node processing times in milliseconds.

5. Compare Performance Results

Repeat the standardized test three times per device to calculate an average execution time:

  • Baseline (CPU): Turn off Use accelerator (OpenCL) under Preferences and run the test.
  • GPU Device 1 (e.g., Integrated Graphics): Re-enable OpenCL, set the system target to the primary GPU, and execute the test.
  • GPU Device 2 (e.g., Dedicated Discrete Card): Change the targeted GPU and repeat the test under the exact same resolution and filter settings.

Shorter processing durations and lower sustained CPU utilization on the GIMP Dashboard confirm the computational throughput and efficiency of the active OpenCL device.