NVIDIA Unveils New Version of CUDA Software Development Tools

NVIDIA is demonstrating the latest version of its award winning C-compiler for GPU programming, CUDA 1.1, at the SuperComputing 2007 show. This latest version of the CUDA software suite includes support for 64-bit Windows XP and starting with the release of CUDA 1.1, NVIDIA's standard display drivers will include the CUDA driver, eliminating the need to install specific software and enabling the seamless deployment of CUDA accelerated applications across multiple markets.

CUDA 1.1 for Windows is available today for free download from www.nvidia.com/cuda. CUDA 1.1 for Linux will be available Monday November 19.

Version 1.1 of the software development kit (SDK) also includes additional multi-GPU C source code examples enabling users to scale CUDA computations across multiple GPUs.

John Stone, senior research programmer at the University of Illinois at Urbana-Champaign (http://www.ks.uiuc.edu/Research/gpu/) commented: "Our scaling efficiency across multiple GPUs is near perfect – four GPUs are delivering almost exactly four-times performance increase in our calculations. A single GPU is delivering 290GFlops, scaling to 1.156 TFlops with four GPUs – this is scaling efficiency of around 99.7 percent."

Also at SuperComputing 07, NVIDIA is previewing a new CUDA debugger based on a standard GDB interface. The debugger enables the setting of break points, as well as setting/examining variables directly on the GPU, all through the C-language. The debugger also supports a standard DDD graphical interface and will allow popular third-party debuggers to seamlessly layer on top of CUDA.