Which GPUs does llama.cpp support?
llama.cpp master (2026-09-27): 53 of the 65 GPUs tracked here are supported, from the library's own requirements. Checked 2026-09-27.
The rule
- NVIDIA: compute capability 5.0 or higher. CUDA backend. Compute capability 5.0-7.0 needs a CUDA 12 build (CUDA 13 builds start at 7.5). Older cards can use the Vulkan backend.
- AMD: any card whose gfx target a current ROCm release supports; others through a fallback backend. HIP backend on cards ROCm supports; the Vulkan backend works on the rest.
- Intel: supported. SYCL or Vulkan backend.
llama.cpp has several GPU backends, so it runs on almost any card; what varies is which backend. The CUDA backend's default build targets are 50/61/70-virtual (CUDA < 13 only), 75/80-virtual, 86-real, 89-real, 90-virtual and 120a-real (CUDA 12.8+), per ggml-cuda's CMakeLists.txt.
Supported (53)
Partly supported (11)
| GPU | Vendor | Target | Details |
|---|---|---|---|
| K80 | NVIDIA | 3.7 | Vulkan backend only |
| Radeon RX 6650 XT | AMD | gfx1032 | Vulkan backend only |
| Radeon RX 6750 XT | AMD | gfx1031 | Vulkan backend only |
| Radeon RX 6600 | AMD | gfx1032 | Vulkan backend only |
| Radeon RX 6700 XT | AMD | gfx1031 | Vulkan backend only |
| Radeon VII | AMD | gfx906 | needs ROCm 6.3.3 or older for the ROCm backend; otherwise Vulkan backend only |
| Radeon Instinct MI50 | AMD | gfx906 | needs ROCm 6.3.3 or older for the ROCm backend; otherwise Vulkan backend only |
| Radeon Instinct MI60 | AMD | gfx906 | needs ROCm 6.3.3 or older for the ROCm backend; otherwise Vulkan backend only |
| Radeon Instinct MI25 | AMD | gfx900 | Vulkan backend only |
| Radeon Instinct MI6 | AMD | gfx803 | Vulkan backend only |
| Radeon Instinct MI8 | AMD | gfx803 | Vulkan backend only |
Not enough data (1)
| GPU | Vendor | Target | Details |
|---|---|---|---|
| Instinct MI455X | AMD | — | gfx target not recorded |
Sources: github.com/ggml-org/llama.cpp/blob/master/ggml/src/ggml-cuda/CMakeLists.txt, github.com/ggml-org/llama.cpp/blob/master/docs/build.md