cuda10.2

Version:

10.2.89

Category:

tools

Cluster:

Loki

Author / Distributor

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

Description

The NVIDIA CUDA® Toolkit 10.2 provides tools, libraries, and development utilities for creating high-performance GPU-accelerated applications.

This version introduces performance improvements and bug fixes while maintaining compatibility with CUDA 10.x workflows.

Available modules:

• cuda10.2/toolkit/10.2.89 Core development toolkit including nvcc, runtime drivers, and API headers.

• cuda10.2/blas/10.2.89 GPU-accelerated implementation of Basic Linear Algebra Subprograms (cuBLAS).

• cuda10.2/fft/10.2.89 Fast Fourier Transform library for GPU acceleration (cuFFT).

CUDA 10.2 supports:

• GPU architectures: Volta, Turing, Pascal

• Optimized libraries: cuBLAS, cuFFT, cuSPARSE, cuRAND

• Nsight profiling tools (via the toolkit module)

Documentation

• Toolkit: https://docs.nvidia.com/cuda/archive/10.2/

• cuBLAS: https://docs.nvidia.com/cuda/archive/10.2/cublas

• cuFFT: https://docs.nvidia.com/cuda/archive/10.2/cufft

• Programming guide: https://docs.nvidia.com/cuda/archive/10.2/cuda-c-programming-guide/

Examples/Usage

• Load the CUDA 10.2 core toolkit:

$ module load cuda10.2/toolkit/10.2.89

• Compile a basic CUDA program:

$ nvcc -o vectorAdd vectorAdd.cu

• Run your application:

$ ./vectorAdd

• Use cuBLAS or cuFFT:

$ module load cuda10.2/blas/10.2.89
$ module load cuda10.2/fft/10.2.89

• Check environment variables:

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

• Unload the modules:

$ module unload cuda10.2/toolkit/10.2.89
$ module unload cuda10.2/blas/10.2.89
$ module unload cuda10.2/fft/10.2.89

Installation

The CUDA 10.2 Toolkit was installed from NVIDIA’s archive: https://developer.nvidia.com/cuda-10.2-download-archive