GP GPUs unleashed: Nvidia releases CUDA 1.0
Santa Clara (CA) – Programmers can now unlock the general processing power of their Nvidia graphics cards with a new SDK. Nvidia’s Compute Unified Device Architecture (CUDA) toolkit allows programmers to use a C-like language which could give “orders of magnitude” increase in compute power for specialized applications.
In Pictures: Tesla, Nvidia's GP GPU strategy
The SDK is available for various flavors of Linux along with Windows. Sample programming examples - which include Black-Scholes options computations and Monte Carlo simulations - can be downloaded to test out the toolkit.
CUDA uses a runtime driver that interoperates with OpenGL and DirectX drivers from Nvidia. The company promises that the toolkit will help programmers better communicate with the graphics chip and adds that traditional ways of programming GPUs made applications “difficult to write, debug, and optimize”.
Graphics processors have been used in various specialized applications like Folding@Home to provide a substantial increase in computation speed. Nvidia hopes the CUDA toolkit will open up its cards to other applications which could include neural network simulations and image processing.
You can download the CUDA SDK and programming samples at Nvidia’s webpage here.