Skip to content

feat: llama.cpp binaries built with cuda13 support #6979

Description

@Minh141120

Problem Statement

After upgrading to Fedora (latest), the system now ships with CUDA 13.

The current llama.cpp builds packaged inside the application only support CUDA 12.x and 11.x. As a result, GPU acceleration fails to initialize, and the local model execution falls back to CPU or errors out entirely.## Feature Idea

Because CUDA 13 introduces ABI changes, the existing CUDA 12 binaries cannot be used — llama.cpp must be rebuilt against the new CUDA toolkit.

Feature Idea

Please provide llama.cpp binaries compiled with:

  • CUDA 13
  • AVX2 CPU optimizations (default target)
  • Ideally also the usual variants (CUDA-cuBLAS, CPU-only, Metal, etc.)

Metadata

Metadata

Labels

Projects

Status
Done

Milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions