b8391
📦 llama-cppView on GitHub →
🐛 3 fixes🔧 1 symbols
Summary
This release adjusts Vulkan behavior to avoid using the graphics queue by default on certain AMD drivers and small GPUs, making its usage opt-in via an environment variable. It also provides updated binaries for multiple operating systems and architectures.
Migration Steps
- To force the use of the graphics queue when using Vulkan, set the environment variable GGML_VK_ALLOW_GRAPHICS_QUEUE.
🐛 Bug Fixes
- Vulkan: Avoided using the graphics queue by default on non-RADV AMD drivers.
- Vulkan: Avoided using the graphics queue on small GPUs.
- Vulkan: Reenabled the transfer queue when the graphics queue is not used.