AMD are getting ready to launch a whole bunch of DirectX 11 compliant graphics cards and RV870 could appear in late summer. NVIDIA is as you perhaps know a bit behind in the DirectX11 plans, but now that more information on its coming GT300 chip has surfaced we realize why NVIDIA is biding its time. NVIDIA’s next generation graphics circuit is very different from the GPUs of today. GT300 will sport twice as many shader processors, 512, as GT200.
NVIDIA will accomplish by using 16 groups of 32 shader processors, GT200 sports 24 shader processors in each cluster and 10 clusters in total.
The most interesting is how these shaders work though. NVIDIA will launched the market’s first MIMD-based GPU (Multiple Instruction stream, Multiple Data stream) says the source. The result will be a GPU with tremendous parallel power since the technology enables shader processors to work individually with different instructions in different data. A technology that will fit perfectly with the growing GPGPU market where NVIDIA has grown strong through its CUDA concept.
If we only consider the number of shader processors GT300 sports twice the power of GT200, if they can keep the clock frequencies equivalent. In applications optimized for parallel workloads the performance can, in theory, improve 6-15 times over the current GT200 chip.
This is of course very premature information, but if it would turn out to be true it would explain why NVIDIA is unable to keep up with AMD. Once again NVIDIA is focusing on raw power and we can only hope for NVIDIA’s sake that it fan outs better than the last attempt to do so.
Leave a Reply