Model name: Intel(R) Xeon(R) Platinum 8559C Threads: 16 (taskset -c 0-15, cgroup cpu.max 1600000/100000) Build: GCC 14.2.0, cmake 3.31.6, -DGGML_NATIVE=ON (AVX512, VNNI, AMX-INT8) llama.cpp base commit: de5122d87 Host was shared; single runs vary by roughly +-8%.