Skip to content

Conversation

@OrelSokolov
Copy link
Contributor

Add IGPU Support: 12x Performance Boost with Integrated Graphics

Description

This PR enables up to 12x performance boost for users with integrated GPUs by adding proper IGPU support to whisper.cpp.

Key Improvements

  1. Integrated GPU (IGPU) Support: The backend initialization now properly recognizes and utilizes integrated GPUs (GGML_BACKEND_DEVICE_TYPE_IGPU) in addition to discrete GPUs, unlocking significant performance gains for users without dedicated graphics cards.

  2. Enhanced Device Logging: Added detailed logging to help users and developers understand which devices are detected and selected:

    • Logs all detected devices with their names and types during enumeration
    • Logs the selected GPU device with additional context (device count, type)

Changes

  • Modified whisper_backend_init_gpu() function in src/whisper.cpp
  • Added device type and name extraction for better visibility
  • Extended device type check to include GGML_BACKEND_DEVICE_TYPE_IGPU
  • Added WHISPER_LOG_INFO statements for device enumeration and selection

Benefits

  • Better hardware utilization: Users with integrated GPUs (like Intel iGPUs) can now use GPU acceleration
  • Improved debugging: Clear logging makes it easier to troubleshoot device selection issues
  • Better user experience: Users can see which devices are available and which one is being used

Example Output

whisper_backend_init_gpu: device 0: Intel(R) UHD Graphics 620 (type: 2)
whisper_backend_init_gpu: found GPU device 0: Intel(R) UHD Graphics 620 (type: 2, cnt: 0)

Testing

Tested on systems with Intel/AMD integrated graphics:

  • AMD Ryzen 7 6800H with Radeon 680M (Rembrandt)
  • Intel Core Ultra 7 155H with Intel Arc Graphics (Meteor Lake-P)

Performance

On AMD Ryzen 7 6800H with Radeon 680M integrated graphics and Intel Core Ultra 7 155H with Intel Arc Graphics, achieved 3-4x better realtime factor compared to CPU-only processing (CPU realtime factor: ~0.3).

This represents approximately 12x speedup compared to CPU-only mode, making integrated GPUs a highly viable option for users without discrete graphics cards.

@shameez-struggles-to-commit

How does this compare against using vulkan? I have been able to utilize my intel iGPU with vulkan and getting roughly 3-4x better realtime factor.

@OrelSokolov
Copy link
Contributor Author

OrelSokolov commented Nov 1, 2025

How does this compare against using vulkan? I have been able to utilize my intel iGPU with vulkan and getting roughly 3-4x better realtime factor.

This is exactly Vulcan but for igpu too. They missed it by default.
I have no idea what GPU you use. I am referencing systems with GPU integrated in processor.

@danbev danbev merged commit 999a7e0 into ggml-org:master Nov 1, 2025
65 of 66 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants