Stop building for CUDA toolkit < 11.8 (#1605)

* Stop building for CUDA toolkit < 11.8

* Simplify

* Drop sm70 from cu128 build targets to align with pytorch

.github/scripts/build-cuda.sh

@@ -8,11 +8,8 @@ set -xeuo pipefail
 # By default, target Maxwell through Hopper.
 build_capability="50;52;60;61;70;75;80;86;89;90"
 
-# CUDA 11.7: Remove sm89 and sm90
-[[ "${cuda_version}" == 11.7.* ]] && build_capability="50;52;60;61;70;75;80;86"
-
-# CUDA 12.8: Add sm100 and sm120; remove sm50 through sm61
-[[ "${cuda_version}" == 12.8.* ]] && build_capability="70;75;80;86;89;90;100;120"
+# CUDA 12.8: Add sm100 and sm120; remove < sm75 to align with PyTorch 2.7+cu128 minimum
+[[ "${cuda_version}" == 12.8.* ]] && build_capability="75;80;86;89;90;100;120"
 
 [[ "${build_os}" = windows-* ]] && python3 -m pip install ninja
 

.github/workflows/python-package.yml

@@ -40,7 +40,7 @@ jobs:
             arch: aarch64
           - os: ubuntu-22.04 # Temporary. Takes too long, not ready yet.
             arch: aarch64
-    runs-on: ${{ matrix.os }} # One day, we could run them on native agents. Azure supports this now but it's planned only for Q3 2023 for hosted agents
+    runs-on: ${{ matrix.os }}
     steps:
       - uses: actions/checkout@v4
       - name: Setup MSVC
@@ -70,13 +70,13 @@ jobs:
           - windows-latest
         arch: [x86_64, aarch64]
         cuda_version:
-          ["11.7.1", "11.8.0", "12.0.1", "12.1.1", "12.2.2", "12.3.2", "12.4.1", "12.5.1", "12.6.3", "12.8.1"]
+          ["11.8.0", "12.0.1", "12.1.1", "12.2.2", "12.3.2", "12.4.1", "12.5.1", "12.6.3", "12.8.1"]
         exclude:
           - os: windows-latest # This probably requires arm64 Windows agents
             arch: aarch64
           - os: ubuntu-22.04 # Temporary. Takes too long, not ready yet.
             arch: aarch64
-    runs-on: ${{ matrix.os }} # One day, we could run them on native agents. Azure supports this now but it's planned only for Q3 2023 for hosted agents
+    runs-on: ${{ matrix.os }}
     steps:
       - uses: actions/checkout@v4
         # Linux: We use Docker to build cross platform Cuda (aarch64 is built in emulation)
@@ -250,33 +250,3 @@ jobs:
         uses: pypa/gh-action-pypi-publish@release/v1
         with:
           print-hash: true
-
-#  test:
-#    needs:
-#      - build-wheels
-#    strategy:
-#      fail-fast: false
-#      matrix:
-#        include:
-#          - os: ubuntu-latest
-#            arch: x86_64
-#            python-version: "3.8"
-#          - os: windows-latest
-#            arch: x86_64
-#            python-version: "3.8"
-#    runs-on: ${{ matrix.os }}
-#    steps:
-#      - uses: actions/checkout@v4
-#      - uses: actions/download-artifact@v4
-#        with:
-#          merge-multiple: true
-#          pattern: "bdist_wheel_${{ matrix.os }}_${{ matrix.arch }}*"
-#          path: wheel/
-#      - uses: actions/setup-python@v5
-#        with:
-#          python-version: ${{ matrix.python-version }}
-#          cache: pip
-#      - shell: bash
-#        run: ls -lar wheel/
-#      - run: pip install wheel/*.whl -r requirements-ci.txt
-#      - run: pytest --log-cli-level=DEBUG --continue-on-collection-errors tests

CMakeLists.txt

@@ -93,8 +93,8 @@ if(BUILD_CUDA)
         )
     endif()
 
-    if(CMAKE_CUDA_COMPILER_VERSION VERSION_LESS "11.0")
-        message(FATAL_ERROR "CUDA Version < 11 is not supported")
+    if(CMAKE_CUDA_COMPILER_VERSION VERSION_LESS "11.4")
+        message(FATAL_ERROR "CUDA Version < 11.4 is not supported")
     elseif(CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL "13.0")
         message(FATAL_ERROR "CUDA Version > 12 is not supported")
     endif()
@@ -103,35 +103,20 @@ if(BUILD_CUDA)
     if(CMAKE_VERSION VERSION_LESS "3.23.0")
         message(STATUS "CMake < 3.23.0; determining CUDA architectures supported...")
 
-        # 11.x and 12.x both support these at a minimum.
-        set(CMAKE_CUDA_ARCHITECTURES_ALL 50 52 53 60 61 62 70 72 75 80)
+        # 11.4+ supports these at a minimum.
+        set(CMAKE_CUDA_ARCHITECTURES_ALL 50 52 53 60 61 62 70 72 75 80 86 87)
         set(CMAKE_CUDA_ARCHITECTURES_ALL_MAJOR 50 60 70 80)
 
-        # CUDA 11.1 adds Ampere support for GA102-GA107.
-        if (CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL "11.1")
-            list(APPEND CMAKE_CUDA_ARCHITECTURES_ALL 86)
-        endif()
-
-        # CUDA 11.4 adds Ampere support for GA10B.
-        if (CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL "11.4")
-            list(APPEND CMAKE_CUDA_ARCHITECTURES_ALL 87)
-        endif()
-
         # CUDA 11.8 adds support for Ada and Hopper.
         if (CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL "11.8")
             list(APPEND CMAKE_CUDA_ARCHITECTURES_ALL 89 90)
             list(APPEND CMAKE_CUDA_ARCHITECTURES_ALL_MAJOR 90)
         endif()
-        # CUDA 12.7 adds support for Blackwell B100.
-        if (CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL "12.7")
-            list(APPEND CMAKE_CUDA_ARCHITECTURES_ALL 100)
-            list(APPEND CMAKE_CUDA_ARCHITECTURES_ALL_MAJOR 100)
-        endif()
 
-        # CUDA 12.8 adds support for RTX 50 Blackwell.
+        # CUDA 12.8 adds support for Blackwell.
         if (CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL "12.8")
-            list(APPEND CMAKE_CUDA_ARCHITECTURES_ALL 101 120)
-            list(APPEND CMAKE_CUDA_ARCHITECTURES_ALL_MAJOR 101 120)
+            list(APPEND CMAKE_CUDA_ARCHITECTURES_ALL 100 101 120)
+            list(APPEND CMAKE_CUDA_ARCHITECTURES_ALL_MAJOR 100 120)
         endif()
     endif()
 

docs/source/installation.mdx

@@ -27,9 +27,9 @@ The latest version of the distributed `bitsandbytes` package is built with the f
 
 | **OS**      | **CUDA Toolkit** | **Host Compiler**         |
 |-------------|------------------|----------------------|
-| **Linux**   | 11.7 - 12.3      | GCC 11.4             |
+| **Linux**   | 11.8 - 12.3      | GCC 11.4             |
 |             | 12.4 - 12.8      | GCC 13.2             |
-| **Windows** | 11.7 - 12.8      | MSVC 19.42+ (VS2022) |
+| **Windows** | 11.8 - 12.8      | MSVC 19.42+ (VS2022) |
 
 For CUDA systems, ensure your hardware meets the following requirements:
 
@@ -88,17 +88,7 @@ For example, to install a compiler and CMake on Ubuntu:
 apt-get install -y build-essential cmake
 ```
 
-You should also install CUDA Toolkit by following the [NVIDIA CUDA Installation Guide for Linux](https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html) guide from NVIDIA. The current expected CUDA Toolkit version is **11.1+** and it is recommended to install **GCC >= 7.3** and required to have at least **GCC >= 6**.
-
-Refer to the following table if you're using another CUDA Toolkit version.
-
-| CUDA Toolkit |  GCC  |
-|--------------|-------|
-| >= 11.4.1    | >= 11 |
-| >= 12.0      | >= 12 |
-| >= 12.4      | >= 13 |
-
-Now to install the bitsandbytes package from source, run the following commands:
+You should also install CUDA Toolkit by following the [NVIDIA CUDA Installation Guide for Linux](https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html) guide from NVIDIA. The current minimum supported CUDA Toolkit version is **11.8**.
 
 ```bash
 git clone https://github.com/bitsandbytes-foundation/bitsandbytes.git && cd bitsandbytes/
@@ -115,13 +105,7 @@ pip install -e .   # `-e` for "editable" install, when developing BNB (otherwise
 
 Windows systems require Visual Studio with C++ support as well as an installation of the CUDA SDK.
 
-To compile from source, you need CMake >= **3.22.1** and Python >= **3.9** installed. You should also install CUDA Toolkit by following the [CUDA Installation Guide for Windows](https://docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/index.html) guide from NVIDIA.
-
-Refer to the following table if you're using another CUDA Toolkit version.
-
-| CUDA Toolkit | MSVC |
-|---|---|
-| >= 11.6 | 19.30+ (VS2022) |
+To compile from source, you need CMake >= **3.22.1** and Python >= **3.9** installed. You should also install CUDA Toolkit by following the [CUDA Installation Guide for Windows](https://docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/index.html) guide from NVIDIA. The current minimum supported CUDA Toolkit version is **11.8**.
 
 ```bash
 git clone https://github.com/bitsandbytes-foundation/bitsandbytes.git && cd bitsandbytes/
@@ -149,7 +133,7 @@ Then locally install the CUDA version you need with this script from bitsandbyte
 ```bash
 wget https://raw.githubusercontent.com/bitsandbytes-foundation/bitsandbytes/main/install_cuda.sh
 # Syntax cuda_install CUDA_VERSION INSTALL_PREFIX EXPORT_TO_BASH
-#   CUDA_VERSION in {110, 111, 112, 113, 114, 115, 116, 117, 118, 120, 121, 122, 123, 124, 125, 126, 128}
+#   CUDA_VERSION in {118, 120, 121, 122, 123, 124, 125, 126, 128}
 #   EXPORT_TO_BASH in {0, 1} with 0=False and 1=True
 
 # For example, the following installs CUDA 12.6 to ~/local/cuda-12.6 and exports the path to your .bashrc

install_cuda.py

@@ -4,14 +4,6 @@ import sys
 from urllib.request import urlretrieve
 
 cuda_versions = {
-    "110": "https://developer.download.nvidia.com/compute/cuda/11.0.3/local_installers/cuda_11.0.3_450.51.06_linux.run",
-    "111": "https://developer.download.nvidia.com/compute/cuda/11.1.1/local_installers/cuda_11.1.1_455.32.00_linux.run",
-    "112": "https://developer.download.nvidia.com/compute/cuda/11.2.2/local_installers/cuda_11.2.2_460.32.03_linux.run",
-    "113": "https://developer.download.nvidia.com/compute/cuda/11.3.1/local_installers/cuda_11.3.1_465.19.01_linux.run",
-    "114": "https://developer.download.nvidia.com/compute/cuda/11.4.4/local_installers/cuda_11.4.4_470.82.01_linux.run",
-    "115": "https://developer.download.nvidia.com/compute/cuda/11.5.2/local_installers/cuda_11.5.2_495.29.05_linux.run",
-    "116": "https://developer.download.nvidia.com/compute/cuda/11.6.2/local_installers/cuda_11.6.2_510.47.03_linux.run",
-    "117": "https://developer.download.nvidia.com/compute/cuda/11.7.1/local_installers/cuda_11.7.1_515.65.01_linux.run",
     "118": "https://developer.download.nvidia.com/compute/cuda/11.8.0/local_installers/cuda_11.8.0_520.61.05_linux.run",
     "120": "https://developer.download.nvidia.com/compute/cuda/12.0.1/local_installers/cuda_12.0.1_525.85.12_linux.run",
     "121": "https://developer.download.nvidia.com/compute/cuda/12.1.1/local_installers/cuda_12.1.1_530.30.02_linux.run",

install_cuda.sh

-URL110=https://developer.download.nvidia.com/compute/cuda/11.0.3/local_installers/cuda_11.0.3_450.51.06_linux.run
-URL111=https://developer.download.nvidia.com/compute/cuda/11.1.1/local_installers/cuda_11.1.1_455.32.00_linux.run
-URL112=https://developer.download.nvidia.com/compute/cuda/11.2.2/local_installers/cuda_11.2.2_460.32.03_linux.run
-URL113=https://developer.download.nvidia.com/compute/cuda/11.3.1/local_installers/cuda_11.3.1_465.19.01_linux.run
-URL114=https://developer.download.nvidia.com/compute/cuda/11.4.4/local_installers/cuda_11.4.4_470.82.01_linux.run
-URL115=https://developer.download.nvidia.com/compute/cuda/11.5.2/local_installers/cuda_11.5.2_495.29.05_linux.run
-URL116=https://developer.download.nvidia.com/compute/cuda/11.6.2/local_installers/cuda_11.6.2_510.47.03_linux.run
-URL117=https://developer.download.nvidia.com/compute/cuda/11.7.1/local_installers/cuda_11.7.1_515.65.01_linux.run
 URL118=https://developer.download.nvidia.com/compute/cuda/11.8.0/local_installers/cuda_11.8.0_520.61.05_linux.run
 URL120=https://developer.download.nvidia.com/compute/cuda/12.0.1/local_installers/cuda_12.0.1_525.85.12_linux.run
 URL121=https://developer.download.nvidia.com/compute/cuda/12.1.1/local_installers/cuda_12.1.1_530.30.02_linux.run
@@ -20,31 +12,7 @@ BASE_PATH=$2
 EXPORT_BASHRC=$3
 
 if [[ -n "$CUDA_VERSION" ]]; then
-  if   [[ "$CUDA_VERSION" -eq "110" ]]; then
-    URL=$URL110
-    FOLDER=cuda-11.0
-  elif [[ "$CUDA_VERSION" -eq "111" ]]; then
-    URL=$URL111
-    FOLDER=cuda-11.1
-  elif [[ "$CUDA_VERSION" -eq "112" ]]; then
-    URL=$URL112
-    FOLDER=cuda-11.2
-  elif [[ "$CUDA_VERSION" -eq "113" ]]; then
-    URL=$URL113
-    FOLDER=cuda-11.3
-  elif [[ "$CUDA_VERSION" -eq "114" ]]; then
-    URL=$URL114
-    FOLDER=cuda-11.4
-  elif [[ "$CUDA_VERSION" -eq "115" ]]; then
-    URL=$URL115
-    FOLDER=cuda-11.5
-  elif [[ "$CUDA_VERSION" -eq "116" ]]; then
-    URL=$URL116
-    FOLDER=cuda-11.6
-  elif [[ "$CUDA_VERSION" -eq "117" ]]; then
-    URL=$URL117
-    FOLDER=cuda-11.7
-  elif [[ "$CUDA_VERSION" -eq "118" ]]; then
+  if [[ "$CUDA_VERSION" -eq "118" ]]; then
     URL=$URL118
     FOLDER=cuda-11.8
   elif [[ "$CUDA_VERSION" -eq "120" ]]; then
@@ -69,10 +37,10 @@ if [[ -n "$CUDA_VERSION" ]]; then
     URL=$URL126
     FOLDER=cuda-12.6
   else
-    echo "argument error: No cuda version passed as input. Choose among versions 110 to 126"
+    echo "argument error: No cuda version passed as input. Choose among versions 118 to 126"
   fi
 else
-    echo "argument error: No cuda version passed as input. Choose among versions 110 to 126"
+    echo "argument error: No cuda version passed as input. Choose among versions 118 to 126"
 fi
 
 FILE=$(basename $URL)