Git commit
7b8443a
Operating systems
Linux
GGML backends
Vulkan
Problem description & steps to reproduce
Hardware Context (Add this in the description)
CPU: AMD Ryzen (Zen 2 / Matisse)
Note: Since this is a Zen 2 CPU (no AVX-512), I used a symlink from libggml-cpu-haswell.so to libggml-cpu.so to ensure compatibility with AVX2.
GPU: AMD Radeon RX 9070 XT (RDNA4 / gfx1201)
Problem description & steps to reproduce
(Translated from Japanese)
I am using the pre-built Vulkan binaries (build 8966) on a fresh install of Ubuntu 24.04.
The program fails to initialize the CPU backend even when offloading all layers to the GPU with
Steps to reproduce:
Download pre-built Vulkan binaries (b8966).
Run ./llama-bench or ./llama-server on Ubuntu 24.04.
The application crashes with: main: error: CPU backend is not loaded.
Investigation:
Running ldd on the backend shared library revealed a missing dependency:
libgomp.so.1 => not found
Solution found:
Installing the OpenMP runtime (sudo apt install libgomp1) resolved the issue.
First Bad Commit
No response
Compile command
N/A (Using pre-built binaries: llama-b8966-bin-ubuntu-vulkan-x64.tar.gz from the official releases page)
Relevant log output
ggml_vulkan: Found 1 Vulkan devices:
ggml_vulkan: 0 = AMD Radeon RX 9070 XT (RADV GFX1201) (radv) [cite: 7, 12, 77]
load_backend: loaded Vulkan backend from /app/llama-b8966/libggml-vulkan.so [cite: 64]
main: error: CPU backend is not loaded
Git commit
7b8443a
Operating systems
Linux
GGML backends
Vulkan
Problem description & steps to reproduce
Hardware Context (Add this in the description)
CPU: AMD Ryzen (Zen 2 / Matisse)
Note: Since this is a Zen 2 CPU (no AVX-512), I used a symlink from libggml-cpu-haswell.so to libggml-cpu.so to ensure compatibility with AVX2.
GPU: AMD Radeon RX 9070 XT (RDNA4 / gfx1201)
Problem description & steps to reproduce
(Translated from Japanese)
I am using the pre-built Vulkan binaries (build 8966) on a fresh install of Ubuntu 24.04.
The program fails to initialize the CPU backend even when offloading all layers to the GPU with
Steps to reproduce:
Download pre-built Vulkan binaries (b8966).
Run ./llama-bench or ./llama-server on Ubuntu 24.04.
The application crashes with: main: error: CPU backend is not loaded.
Investigation:
Running ldd on the backend shared library revealed a missing dependency:
libgomp.so.1 => not found
Solution found:
Installing the OpenMP runtime (sudo apt install libgomp1) resolved the issue.
First Bad Commit
No response
Compile command
Relevant log output