BCS Blog

COBOL, z/OS performance measurement, and systems programming updates.

OneAPI

By Arch Brooks on July 10, 2026 in Programming

Back to Blog ← Back

Permalink

The fastest method on Gentoo is to build the current upstream version with CMake.

Install dependencies

sudo emerge --ask \
    git \
    cmake \
    ninja \
    clang \
    llvm \
    openmp \
    blas \
    lapack

If you have already installed Intel oneAPI (which you've mentioned previously), source it:

source /opt/intel/oneapi/setvars.sh

Clone llama.cpp

cd ~/Applications
git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp

Build (CPU)

cmake -B build \
    -G Ninja \
    -DCMAKE_BUILD_TYPE=Release

cmake --build build -j$(nproc)

Build (Intel Arc GPU using SYCL)

source /opt/intel/oneapi/setvars.sh

cmake -B build \
    -G Ninja \
    -DCMAKE_BUILD_TYPE=Release \
    -DGGML_SYCL=ON \
    -DCMAKE_C_COMPILER=icx \
    -DCMAKE_CXX_COMPILER=icpx

cmake --build build -j$(nproc)

Verify

build/bin/llama-cli --version

build/bin/llama-bench

Add to PATH

echo 'export PATH="$HOME/Applications/llama.cpp/build/bin:$PATH"' >> ~/.bashrc
source ~/.bashrc

Verify:

which llama-cli

Test a model

mkdir -p ~/Models

Place a GGUF model into ~/Models, then run:

llama-cli \
    -m ~/Models/qwen3-8b-q4_k_m.gguf \
    -ngl 999 \
    -c 8192

Confirm the Intel Arc GPU is being used

sycl-ls

Then run:

llama-bench \
    -m ~/Models/qwen3-8b-q4_k_m.gguf \
    -ngl 999

You should see output indicating the SYCL backend is in use, for example:

ggml_backend_init: found SYCL device
Intel(R) Arc Graphics

This confirms inference is running on the Intel Arc GPU rather than the CPU.

Comments

No comments yet.

Title *

Slug

Excerpt

Content *

HTML is supported.

SEO Meta Data

Meta Title

Meta Description