@@ -53,9 +53,11 @@ cuNumeric is available [on conda](https://anaconda.org/legate/cunumeric):
5353```
5454conda install -c nvidia -c conda-forge -c legate cunumeric
5555```
56+ The conda package is compatible with CUDA >= 11.4 (CUDA driver version >= r470),
57+ and Volta or later GPU architectures.
5658
57- Pre-built docker images containing all Legate libraries , as well as specialized
58- install scripts for supported clusters are available on the
59+ Docker image build scripts , as well as specialized install scripts for
60+ supported clusters are available on the
5961[ quickstart] ( https://github.com/nv-legate/quickstart ) repo.
6062
6163Read on for general instructions on building cuNumeric from source.
@@ -68,17 +70,25 @@ library prior to installing cuNumeric.
6870
6971cuNumeric requires the following:
7072
71- - Python >= 3.7
72- - [ CUDA] ( https://developer.nvidia.com/cuda-downloads ) >= 8.0
73+ - Python >= 3.8
74+ - [ CUDA] ( https://developer.nvidia.com/cuda-downloads ) >= 10.2
7375 - GNU Make
74- - C++14 compatible compiler (g++, clang, or nvc++)
76+ - C++17 compatible compiler (g++, clang, or nvc++)
7577 - Fortran compiler (for building OpenBLAS; not necessary if you provide a pre-built version of OpenBLAS)
76- - the Python packages listed in ` conda/cunumeric_dev.yml `
78+ - the Python packages listed in any one of the conda environment files:
79+ - ` conda/environment-test-3.8.yml `
80+ - ` conda/environment-test-3.9.yml `
81+ - ` conda/environment-test-3.10.yml `
7782
7883See the [ corresponding section] ( https://github.com/nv-legate/legate.core#dependencies )
7984on the Legate Core instructions for help on installing the required Python packages
8085using conda.
8186
87+ cuNumeric is tested and guaranteed to be compatible with Volta and later GPU
88+ architectures. You can use cuNumeric with Pascal GPUs as well, but there could
89+ be issues due to lack of independent thread scheduling. Please report any such
90+ issues on GitHub.
91+
8292## Building from Source
8393
8494Installation of cuNumeric is done with either ` setup.py ` for simple
@@ -92,22 +102,30 @@ python setup.py --with-core <path-to-legate-core-installation>
92102This will build cuNumeric against the Legate Core installation and then
93103install cuNumeric into the same location.
94104
95- Note that after the first invocation of ` setup.py ` this repository will remember
96- which Legate Core installation to use and the ` -- with-core ` option can be
97- omitted unless the user wants to change it.
98-
99- Advanced users can also invoke ` install .py --help ` to see options for
100- configuring cuNumeric by invoking the ` install.py ` script directly.
105+ If Legate Core has been installed with CUDA support, a working cuTENSOR
106+ installation must also be provided to the installation command with the
107+ ` --with-cutensor ` option:
108+ ```
109+ python setup .py --with-core <path-to-legate-core-installation> --with-cutensor <path-to-cutensor-installation>
110+ ```
101111
102- Of particular interest to cuNumeric users will likely be the option for
103- specifying an installation of [ OpenBLAS] ( https://www.openblas.net/ ) to use.
104- If you already have an installation of OpenBLAS on your machine you can
105- inform the ` install.py ` script about its location using the ` --with-openblas ` flag :
112+ You can also specify an installation of [ OpenBLAS ] ( https://www.openblas.net/ )
113+ to use for the build. If you already have an installation of OpenBLAS on your
114+ machine, you can inform the installation script using the ` --with-openblas `
115+ option :
106116
107117```
108- python setup.py --with-openblas / path/to/open/blas/
118+ python setup.py --with-openblas < path-to-OpenBLAS>
109119```
110120
121+ Advanced users can also invoke ` install.py --help ` to see options for
122+ configuring cuNumeric by invoking the ` install.py ` script directly.
123+
124+ Note that after the first invocation of ` setup.py ` or ` install.py ` this
125+ repository will remember the paths you provided with the installation options
126+ and they can be omitted in subsequent invocations unless you want to change
127+ them.
128+
111129## Usage and Execution
112130
113131Using cuNumeric as a replacement for NumPy is easy. Users only need
@@ -131,7 +149,7 @@ legate cunumeric_program.py
131149```
132150
133151For execution with multiple nodes (assuming Legate Core is installed with GASNet support)
134- users can supply the ` --nodes ` flag . For execution with GPUs, users can use the
152+ users can supply the ` --nodes ` option . For execution with GPUs, users can use the
135153` --gpus ` flags to specify the number of GPUs to use per node. We encourage all users
136154to familiarize themselves with these resource flags as described in the Legate Core
137155documentation or simply by passing ` --help ` to the ` legate ` driver script.
0 commit comments