According to those reports, Griffin will be a 65nm dual-core chip with 1MB of L2 cache per core. Unlike AMD's upcoming Barcelona processor, it won't have a shared L3 cache, but it will nonetheless have split power planes between both cores and the north bridge/memory controller. The chip will also be able to scale to 1/8th of its rated clock speed in order to conserve power—for instance, a hypothetical 2.4GHz Griffin processor would be able to clock itself all the way down to 300MHz. Further power savings will be brought by HyperTransport 3.0, which will dynamically scale link widths to save energy. On the memory front, Bit-Tech says Griffin will support DDR2-667 and DDR2-800 memory, and perhaps even DDR2-1066 like Barcelona.
Along with the new Griffin processor, AMD's Puma platform will include a new chipset dubbed 780G. AMD's slides say the 780G will monitor I/O and the processor's power state to establish the "optimal power management profile for a given set of system conditions." The 780G will link a chunk of NAND flash memory to the chipset's south bridge, as well, offering flash caching capability similar to that of Intel's Turbo Memory. The 780G chipset also will support "hybrid graphics"—dynamic switching between integrated and discrete graphics.
The Puma platform is reportedly scheduled to hit production in the fourth quarter, but Puma-based notebooks will only see the light of day next year. CNet quotes AMD Fellow Maurice Steinman as saying we'll actually have to wait until mid-2008 for Puma notebooks.