cuda11.4

Version:: 11.4.2
Category:: tools
Cluster:: Loki

Author / Distributor

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

Description

The NVIDIA CUDA® Toolkit 11.4 delivers tools, libraries, and development utilities for GPU-accelerated computing. It continues support for Ampere, Turing, Volta, and Pascal architectures with optimizations across the toolchain.

Available modules:

cuda11.4/toolkit/11.4.2 Core development toolkit including nvcc, device/runtime drivers, Nsight tools, and headers
cuda11.4/blas/11.4.2 cuBLAS 11.5 — highly optimized GPU-accelerated linear algebra routines
cuda11.4/fft/11.4.2 cuFFT — Fast Fourier Transform library for GPU workloads

Highlights in CUDA 11.4:

Enhanced support for A100 GPUs and multi-instance GPU (MIG) features
Updates to cuBLAS 11.5, cuSPARSE, cuFFT, and other libraries
Improved debugging/profiling via Nsight Systems and Nsight Compute
Expanded CUDA Graph features
Faster device memory initialization via async cudaMemcpy

Documentation

Toolkit: https://docs.nvidia.com/cuda/archive/11.4/
cuBLAS: https://docs.nvidia.com/cuda/archive/11.4/cublas
cuFFT: https://docs.nvidia.com/cuda/archive/11.4/cufft
Programming Guide: https://docs.nvidia.com/cuda/archive/11.4/cuda-c-programming-guide/

Examples/Usage

Load the CUDA 11.4 toolkit module:

$ module load cuda11.4/toolkit/11.4.2

Compile a CUDA application targeting Ampere:

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

Run the compiled program:

$ ./app

Load cuBLAS or cuFFT separately if needed:

$ module load cuda11.4/blas/11.4.2
$ module load cuda11.4/fft/11.4.2

Inspect your environment:

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

Unload modules:

$ module unload cuda11.4/toolkit/11.4.2
$ module unload cuda11.4/blas/11.4.2
$ module unload cuda11.4/fft/11.4.2

Installation

The CUDA 11.4 Toolkit was installed from: https://developer.nvidia.com/cuda-11.4-download-archive