Sigmadelt

Despite the ratification of OpenCL as a standard may have changed significantly in the scenario of GPGPU applications, Nvidia has no intention of abandoning Cuda, a platform on which it has invested and will continue to invest.

The American company has decided to present a new version, for now beta, release 2.1, which introduces support for Windows Vista, for products Tesla and a 32-bit debugger can run on RedHat Enterprise Linux 5.x. It also added support for VisualStudio 2008 on Windows XP. According to the company were included in the new API can work properly with Direct3D 9 and Direct3D 10, this must be added a number of improvements in interoperability with OpenGL in addition to support for many of the recent Linux distributions.

Nvidia, as AMD has recently announced that developers can use the SDK to create applications based on OpenCL able to exploit the GPU. The Californian company had, the beginning of the year, also announced the forthcoming introduction of a platform capable of Cuda function properly on x86 systems: the launch was subsequently postponed to the end of 2008, and now, according to many sources, such platform may not ever arrive. In this scenario it should be added to Intel’s Larrabee, a product that American society has long since dropped the curtain, preferring to work behind the scenes. It predicts a hot and intense in 2009 as regards the reality of the applications with accelerated graphics card.

NVIDIA is preparing to present a new set of drivers with support for the PhysX API, through the use of their GPU with unified shader architecture and Cuda.

Underlying these drivers are of course the acquisition of Ageî by NVIDIA, which occurred in recent months and that will soon take the first concrete in the form in fact a driver that provides compatibility between these APIs and GPU NVIDIA Series GeForce 8, GeForce 9, and the next-generation GeForce GTX 200.

The PhysX API is particularly widespread in both the console and in the PC, allowing you to implement the physical effects of 3D objects. Until now it was possible to exploit these APIs with the PhysX hardware solutions, the so-called PPU (Physics Processing Unit) developed by Ageî, but following the acquisition by NVIDIA all research and development efforts will go in the direction of using the GPU for calculations related to physics.

The unique feature of the PhysX API is currently not implemented via Cuda GPU is in the management of rigid bodies, it is not clear at present moment if and when this function can be somewhat accelerated by the GPU.

The processing power of recent GPUs with NVIDIA physical processes linked to is much higher than that made available by the PPU Ageî on the market. On the other hand, the GPU must provide for all the processes related to traditional 3D rendering, so only a part the resources available can be used for the management of physics.

This obviously opens up space for hybrid-type approaches, in which one or two GPUs are used for the traditional rendering 3D GPU while another is specifically designed for use with PhysX API in order to obtain the highest possible acceleration of the effects related to physics.