NVIDIA cuRAND is a high-performance GPU-accelerated random number generation (RNG) library. It delivers high-quality random numbers up to 8x faster by utilizing the hundreds of processor cores available in NVIDIA GPUs. The cuRAND library is included in both the NVIDIA HPC SDK and the CUDA Toolkit, making it accessible for a wide range of applications.

Features

• Flexible Usage Model: Offers a host API for generating random numbers in bulk on the GPU and an inline implementation for use inside GPU functions/kernels or in host code.
• High-Quality RNG Algorithms: Includes four high-quality RNG algorithms—MRG32k3a, MTGP (Mersenne Twister for Graphics Processors), XORWOW, and Sobol’ quasi-random number generators, with support for scrambled and 64-bit RNG.
• Multiple RNG Distribution Options: Supports uniform, normal, log-normal, single-precision or double-precision, and Poisson distributions.
• Tested Against Statistical Test Batteries: The random number generators and statistical distributions have been tested against well-known statistical test batteries, including TestU01.

Benefits

• High Performance: Delivers high-quality random numbers up to 8x faster using NVIDIA GPUs.
• Versatility: Supports a variety of RNG algorithms and distribution options, making it suitable for diverse applications.
• Flexibility: Provides both host and device APIs, allowing for bulk generation and inline implementation within CUDA functions/kernels.
• Reliability: Tested against well-known statistical test batteries to ensure the quality and reliability of generated random numbers.