Merge pull request #8129 from leofang/core_doc

Update conda installation guide

README.md

@@ -57,7 +57,9 @@ Binary packages are also available for Linux and Windows on [Conda-Forge](https:
 | --------------------- | --------------------------- | ------------------------------------------------------------- |
 | CUDA                  | x86_64 / aarch64 / ppc64le  | `conda install -c conda-forge cupy`                           |
 
-If you need to use a particular CUDA version (say 11.8), you can do `conda install -c conda-forge cupy cuda-version=11.8`.
+If you need a slim installation (without also getting CUDA dependencies installed), you can do `conda install -c conda-forge cupy-core`.
+
+If you need to use a particular CUDA version (say 12.0), you can use the `cuda-version` metapackage to select the version, e.g. `conda install -c conda-forge cupy cuda-version=12.0`.
 
 > [!NOTE]\
 > If you encounter any problem with CuPy installed from `conda-forge`, please feel free to report to [cupy-feedstock](https://github.com/conda-forge/cupy-feedstock/issues), and we will help investigate if it is just a packaging issue in `conda-forge`'s recipe or a real issue in CuPy.

docs/source/install.rst

@@ -110,17 +110,24 @@ Please make sure that only one CuPy package (``cupy`` or ``cupy-cudaXX`` where X
 Installing CuPy from Conda-Forge
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
-Conda/Anaconda is a cross-platform package management solution widely used in scientific computing and other fields.
+Conda is a cross-language, cross-platform package management solution widely used in scientific computing and other fields.
 The above ``pip install`` instruction is compatible with ``conda`` environments. Alternatively, for both Linux (x86_64,
 ppc64le, aarch64-sbsa) and
 Windows once the CUDA driver is correctly set up, you can also install CuPy from the ``conda-forge`` channel::
 
     $ conda install -c conda-forge cupy
 
 and ``conda`` will install a pre-built CuPy binary package for you, along with the CUDA runtime libraries
-(``cudatoolkit``). It is not necessary to install CUDA Toolkit in advance.
+(``cudatoolkit`` for CUDA 11 and below, or ``cuda-XXXXX`` for CUDA 12 and above). It is not necessary to install CUDA Toolkit in advance.
 
-Conda has a built-in mechanism to determine and install the latest version of ``cudatoolkit`` supported by your driver.
+If you aim at minimizing the installation footprint, you can install the ``cupy-core`` package::
+
+    $ conda install -c conda-forge cupy-core
+
+which only depends on ``numpy``. None of the CUDA libraries will be installed this way, and it is your responsibility to install the needed
+dependencies youself, either from conda-forge or elsewhere. This is equivalent of the ``cupy-cudaXX`` wheel installation.
+
+Conda has a built-in mechanism to determine and install the latest version of ``cudatoolkit`` or any other CUDA components supported by your driver.
 However, if for any reason you need to force-install a particular CUDA version (say 11.8), you can do::
 
     $ conda install -c conda-forge cupy cuda-version=11.8
@@ -141,10 +148,12 @@ However, if for any reason you need to force-install a particular CUDA version (
 
 .. note::
 
-    If you did not install CUDA Toolkit by yourself, the ``nvcc`` compiler might not be available, as
+    If you did not install CUDA Toolkit by yourself, for CUDA 11 and below the ``nvcc`` compiler might not be available, as
     the ``cudatoolkit`` package from ``conda-forge`` does not include the ``nvcc`` compiler toolchain. If you would like to use
     it from a local CUDA installation, you need to make sure the version of CUDA Toolkit matches that of ``cudatoolkit`` to
-    avoid surprises.
+    avoid surprises. For CUDA 12 and above, ``nvcc`` can be installed on a per-``conda`` environment basis via
+
+        $ conda install -c conda-forge cuda-nvcc
 
 
 .. _install_cupy_from_source: