We’re happy to announce the release of the NVIDIA CUDA SDK version 0.8.1 beta. Note that this is a new CUDA SDK (i.e. code samples and documentation), and not a new CUDA Toolkit (compiler, runtime, CUBLAS, CUFFT, etc.). You should continue using the NVIDIA CUDA Toolkit version 0.8 beta with this new CUDA SDK.
The new SDK for Linux and Windows is available here:
The CUDA SDK (0.8.1 beta) is now a separate download from the CUDA Toolkit (0.8 beta). Also, we have added a web page where you can view and download individual CUDA code samples and white papers:
The NVIDIA CUDA SDK 0.8.1 contains 29 code samples (27 on Linux). New samples in this release include computational Finance examples like Black-Scholes and Binomial option pricing; fluid simulation using CUFFT (OpenGL and Direct3D 9 versions); and image processing samples like convolution, box filter, DXTC texture compression, and OpenGL scene post-processing.
We have also provided a multi-GPU code sample, a bandwidth test sample that demonstrates how to achieve over 3GB/second between the host and GPU using page-locked memory allocation, and an example of efficiently loading aligned custom data types. Finally, the transpose sample has been improved to demonstrate how to efficiently transpose matrices of arbitrary dimensions.
We hope you enjoy this update. Please use the forums for questions and to report problems.
The NVIDIA CUDA Team