iso: improve burn-in, export, and live boot

This commit is contained in:
Mikhail Chusavitin
2026-03-26 18:56:19 +03:00
parent 67a215c66f
commit fc5c2019aa
23 changed files with 1706 additions and 168 deletions

View File

@@ -159,6 +159,16 @@ else
echo "=== bee binary up to date, skipping build ==="
fi
echo ""
echo "=== downloading cuBLAS/cuBLASLt/cudart ${NCCL_CUDA_VERSION} userspace ==="
sh "${BUILDER_DIR}/build-cublas.sh" \
"${CUBLAS_VERSION}" \
"${CUDA_USERSPACE_VERSION}" \
"${NCCL_CUDA_VERSION}" \
"${DIST_DIR}"
CUBLAS_CACHE="${DIST_DIR}/cublas-${CUBLAS_VERSION}+cuda${NCCL_CUDA_VERSION}"
GPU_STRESS_NEED_BUILD=1
if [ -f "$GPU_STRESS_BIN" ] && [ "${BUILDER_DIR}/bee-gpu-stress.c" -ot "$GPU_STRESS_BIN" ]; then
GPU_STRESS_NEED_BUILD=0
@@ -167,6 +177,7 @@ fi
if [ "$GPU_STRESS_NEED_BUILD" = "1" ]; then
echo "=== building bee-gpu-stress ==="
gcc -O2 -s -Wall -Wextra \
-I"${CUBLAS_CACHE}/include" \
-o "$GPU_STRESS_BIN" \
"${BUILDER_DIR}/bee-gpu-stress.c" \
-ldl
@@ -283,6 +294,10 @@ NCCL_CACHE="${DIST_DIR}/nccl-${NCCL_VERSION}+cuda${NCCL_CUDA_VERSION}"
cp "${NCCL_CACHE}/lib/"* "${OVERLAY_STAGE_DIR}/usr/lib/"
echo "=== NCCL: $(ls "${NCCL_CACHE}/lib/" | wc -l) files injected into /usr/lib/ ==="
# Inject cuBLAS/cuBLASLt/cudart runtime libs used by bee-gpu-stress tensor-core GEMM path
cp "${CUBLAS_CACHE}/lib/"* "${OVERLAY_STAGE_DIR}/usr/lib/"
echo "=== cuBLAS: $(ls "${CUBLAS_CACHE}/lib/" | wc -l) files injected into /usr/lib/ ==="
# --- embed build metadata ---
mkdir -p "${OVERLAY_STAGE_DIR}/etc"
BUILD_DATE="$(date +%Y-%m-%d)"
@@ -297,6 +312,8 @@ DEBIAN_KERNEL_ABI=${DEBIAN_KERNEL_ABI}
NVIDIA_DRIVER_VERSION=${NVIDIA_DRIVER_VERSION}
NCCL_VERSION=${NCCL_VERSION}
NCCL_CUDA_VERSION=${NCCL_CUDA_VERSION}
CUBLAS_VERSION=${CUBLAS_VERSION}
CUDA_USERSPACE_VERSION=${CUDA_USERSPACE_VERSION}
EOF
# Patch motd with build info