Expected to start with a 700MHz GPU clock, 1100MHz memory clock and 1600MHz shader clock, the raw horsepower of the G300 will likely be very high. This will be coupled with the 512 shader units present in the G300, well over twice the number of shader units most cards today have. The high memory clock means that they expect the card to be able to push 281.6GB/s of memory bandwidth, twice as much as the GTX 280.
The G300 will also incorporate new stream processors, making use of newer instruction sets that supplant SIMD execution with “MIMD-like” execution, indicating the G300 will be a highly parallel card with an estimated processing potential of up to 2.5 teraFLOPS. Nvidia hasn't announced any breakthroughs in process reduction or manufacturing technique, which means G300 will likely be a 40nm product. I look forward to hearing more accurate info straight from Nvidia as it becomes available.