From 83bbc8a1bcde81ac8eb80c89fb247ca5aa46b1ff Mon Sep 17 00:00:00 2001
From: Michael Chus <mike@mchus.pro>
Date: Thu, 26 Mar 2026 23:51:26 +0300
Subject: [PATCH] fix(nccl-tests): upgrade to cuda-nvcc-12-8, add sm_100
 (Blackwell B100/B200)

---
 iso/builder/Dockerfile          | 2 +-
 iso/builder/VERSIONS            | 2 +-
 iso/builder/build-nccl-tests.sh | 9 +++++----
 3 files changed, 7 insertions(+), 6 deletions(-)

diff --git a/iso/builder/Dockerfile b/iso/builder/Dockerfile
index e485ad4..36a3de6 100644
--- a/iso/builder/Dockerfile
+++ b/iso/builder/Dockerfile
@@ -36,7 +36,7 @@ RUN wget -qO /tmp/cuda-keyring.gpg \
 https://developer.download.nvidia.com/compute/cuda/repos/debian12/x86_64/ /" \
         > /etc/apt/sources.list.d/cuda.list \
     && apt-get update -qq \
-    && apt-get install -y cuda-nvcc-12-6 \
+    && apt-get install -y cuda-nvcc-12-8 \
     && rm -rf /var/lib/apt/lists/*
 
 RUN arch="$(dpkg --print-architecture)" \
diff --git a/iso/builder/VERSIONS b/iso/builder/VERSIONS
index a1e5b50..6c46761 100644
--- a/iso/builder/VERSIONS
+++ b/iso/builder/VERSIONS
@@ -5,7 +5,7 @@ NCCL_VERSION=2.28.9-1
 NCCL_CUDA_VERSION=13.0
 NCCL_SHA256=2e6faafd2c19cffc7738d9283976a3200ea9db9895907f337f0c7e5a25563186
 NCCL_TESTS_VERSION=2.13.10
-NVCC_VERSION=12.6
+NVCC_VERSION=12.8
 CUBLAS_VERSION=13.0.2.14-1
 CUDA_USERSPACE_VERSION=13.0.96-1
 GO_VERSION=1.24.0
diff --git a/iso/builder/build-nccl-tests.sh b/iso/builder/build-nccl-tests.sh
index 7ffe3f3..4a2f90a 100755
--- a/iso/builder/build-nccl-tests.sh
+++ b/iso/builder/build-nccl-tests.sh
@@ -34,9 +34,9 @@ if [ -f "${CACHE_DIR}/bin/all_reduce_perf" ]; then
     exit 0
 fi
 
-# Resolve nvcc path (cuda-nvcc-12-6 installs to /usr/local/cuda-12.6/bin/nvcc)
+# Resolve nvcc path (cuda-nvcc-12-8 installs to /usr/local/cuda-12.8/bin/nvcc)
 NVCC=""
-for candidate in nvcc /usr/local/cuda-12.6/bin/nvcc /usr/local/cuda-12/bin/nvcc /usr/local/cuda/bin/nvcc; do
+for candidate in nvcc /usr/local/cuda-12.8/bin/nvcc /usr/local/cuda-12/bin/nvcc /usr/local/cuda/bin/nvcc; do
     if command -v "$candidate" >/dev/null 2>&1 || [ -x "$candidate" ]; then
         NVCC="$candidate"
         break
@@ -96,11 +96,12 @@ SRC_DIR=$(ls -d nccl-tests-* 2>/dev/null | head -1)
 cd "$SRC_DIR"
 
 echo "=== building all_reduce_perf ==="
-# CUDA 12.6 supports Volta through Hopper (sm_70..sm_90).
+# CUDA 12.8 supports Volta through Blackwell (sm_70..sm_100).
 GENCODE="-gencode=arch=compute_70,code=sm_70 \
          -gencode=arch=compute_80,code=sm_80 \
          -gencode=arch=compute_86,code=sm_86 \
-         -gencode=arch=compute_90,code=sm_90"
+         -gencode=arch=compute_90,code=sm_90 \
+         -gencode=arch=compute_100,code=sm_100"
 make MPI=0 \
     NVCC="$NVCC" \
     CUDA_HOME="$CUDA_HOME" \