cuda11.2

Version:

11.2.0

Category:

tools

Cluster:

Loki

Author / Distributor

https://developer.nvidia.com/cuda-toolkit-archive

Description

The NVIDIA CUDA® Toolkit 11.2 provides a robust development environment for creating GPU-accelerated applications using C, C++, and Fortran.

New in CUDA 11.2:

• Faster compilation with nvcc using Link Time Optimization (LTO)

• Expanded CUDA Graph APIs

• Enhanced support for the Ampere architecture (e.g., A100 GPUs)

• New library versions: - cuBLAS 11.4 - cuFFT 10.4 - cuSPARSE 11.3 - NCCL 2.8+

CUDA 11.2 continues to support:

• Volta, Turing, and Pascal GPUs

• Compatibility with multiple GCC versions (up to GCC 9.x)

Available modules:

• cuda11.2/toolkit/11.2.0 Core compiler (nvcc), CUDA runtime, Nsight tools, and development headers

• cuda11.2/blas/11.2.0 cuBLAS 11.4 — optimized linear algebra routines for dense matrix operations

• cuda11.2/fft/11.2.0 cuFFT 10.4 — GPU-accelerated Fast Fourier Transform routines

Documentation

• Toolkit: https://docs.nvidia.com/cuda/archive/11.2/

• cuBLAS: https://docs.nvidia.com/cuda/archive/11.2/cublas

• cuFFT: https://docs.nvidia.com/cuda/archive/11.2/cufft

• Programming Guide: https://docs.nvidia.com/cuda/archive/11.2/cuda-c-programming-guide/

Examples/Usage

• Load the core CUDA module:

$ module load cuda11.2/toolkit/11.2.0

• Compile a program targeting Ampere:

$ nvcc -arch=sm_80 my_kernel.cu -o my_kernel

• Run the program:

$ ./my_kernel

• Load additional libraries if needed:

$ module load cuda11.2/blas/11.2.0
$ module load cuda11.2/fft/11.2.0

• Check your environment configuration:

$ echo $CUDA_HOME
$ echo $PATH
$ echo $LD_LIBRARY_PATH

• Unload all modules:

$ module unload cuda11.2/toolkit/11.2.0
$ module unload cuda11.2/blas/11.2.0
$ module unload cuda11.2/fft/11.2.0

Installation

The CUDA 11.2 Toolkit was installed from: https://developer.nvidia.com/cuda-11.2-download-archive