Graphics cards have been based on SIMD (Single-Instruction Multiple Data) architecture since the first graphical processor was launched. However, NVIDIA is going to change that by introducing GT300 later this year. GT300 will be based on completely new MIMD (Multiple-Instruction Multiple Data) architecture. The new MIMD architecture will not only improve graphical performance of the chip, but also general computing performance of the GPU.

The GT300 will have 16 groups of 32 cores resulting in total of 512 processing units. This number will by itself result in 2X performance when compared to GT200 chip. A core clock speed of 2GHz will result in 3 TFLOPS performance and with new MIMD architecture the Dual precision performance should also see 6-15x improvement over GT200.

NVIDIA has also improved the Scratch Cache of the GT300 by making it more granular, which would allow more interactivity between the cores inside the cluster. General computing on GT300 will see vast improvement as the new chip will allow direct hardware access to programming environments like CUDA 3.0, DirectX 11, OpenGL 3.0 and OpenCL. It is believed that GT300 will surprise everyone with its performance in Single and Dual precision data, making it the fastest GPU available for both graphical and general computing purposes.
[Via Bright Side of News]