With the first of NVIDIA's GeForce 400 "Fermi" graphics cards arriving later this month, their software engineers have put out the release of CUDA 3.0. Version 3 of the Compute Unified Device Architecture has a wealth of changes including Fermi support, C++ support, a new unified interoperability API for Direct3D (including Direct3D 11.0) and OpenGL (including OpenGL 3.x/4.0), up to a 100x performance increase when debugging with cuda-gdb, a new CUDA memory checker, and support for all the OpenCL features in the latest R195 production driver package.
The NVIDIA Fermi support in CUDA 3.0 includes native 64-bit GPU support, multiple copy engine support, ECC reporting, concurrent kernel execution, Fermi hardware debugging support via cuda-gdb, and Fermi hardware profiling support for CUDA's C and OpenCL via the NVIDIA Visual Profiler. The first of the NVIDIA Fermi graphics cards in the GeForce 400 series, the NVIDIA GeForce GTX 470 and GeForce GTX 480, are expected to be launched in just a few days and are both built upon the "GF100" core.
Details and downloads on the NVIDIA CUDA 3.0 software can be found at NVIDIA.com
. The CUDA Toolkit is officially available on Linux for Fedora 10, Red Hat Enterprise Linux 4.8, Red Hat Enterprise Linux 5.3, Ubuntu Linux 9.04, openSUSE 11.1, and SUSE Linux Enterprise Desktop 11.