cuda11.1

Version:

11.1.1

Category:

tools

Cluster:

Loki

Author / Distributor

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

Description

The NVIDIA CUDA® Toolkit 11.1 provides the tools, compiler, and GPU-accelerated libraries necessary to build high-performance applications across supported NVIDIA GPUs.

This version extends CUDA 11.0 with performance improvements, Ampere architecture support, and updated libraries.

Available modules:

• cuda11.1/toolkit/11.1.1 Core compiler and development tools (nvcc, headers, runtime, and drivers)

• cuda11.1/blas/11.1.1 cuBLAS 11.3 — optimized dense linear algebra routines for GPUs

• cuda11.1/fft/11.1.1 cuFFT — Fast Fourier Transform library for high-performance signal and frequency domain computation

• Support for Ampere architecture (e.g., GA100 / A100)

• Enhanced CUDA Graphs support

• Updated Nsight Compute 2020.2 and Nsight Systems 2020.3

• Compatibility with newer cuDNN and NCCL versions

Documentation

• Toolkit: https://docs.nvidia.com/cuda/archive/11.1/

• cuBLAS: https://docs.nvidia.com/cuda/archive/11.1/cublas

• cuFFT: https://docs.nvidia.com/cuda/archive/11.1/cufft

• CUDA Programming Guide: https://docs.nvidia.com/cuda/archive/11.1/cuda-c-programming-guide/

Examples/Usage

• Load the CUDA Toolkit:

$ module load cuda11.1/toolkit/11.1.1

• Compile your application (for Ampere GPU):

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

• Run your executable:

$ ./my_code

• Load cuBLAS or cuFFT if needed explicitly:

$ module load cuda11.1/blas/11.1.1
$ module load cuda11.1/fft/11.1.1

• Check environment paths:

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

• Unload modules when done:

$ module unload cuda11.1/toolkit/11.1.1
$ module unload cuda11.1/blas/11.1.1
$ module unload cuda11.1/fft/11.1.1

Installation

The CUDA 11.1 Toolkit was downloaded from the NVIDIA archive: https://developer.nvidia.com/cuda-11.1-download-archive