Back to feed

b8308

Mar 13, 2026
Meta/llama.cppCLIvb8308

vulkan: Fix ErrorOutOfHostMemory on Intel GPU when loading large models with --no-mmap (#20059)

  • Changed to reuse command buffers to fix crashing on Intel GPU

  • Removed unused parameter

  • Fixed compile error and minor mistake

  • Fix logging

  • Changing to use usage flag per command buffer

  • fixed style

  • added buffer reset

  • Removed cmd_buffer_idx for reuse consistency

  • Fixed style

macOS/iOS:

Linux:

Windows:

openEuler: