P4:
memory -->up to 6.4 GB/sec--> memory controller/northbridge -->up to 6.4GB/sec--> CPU
A64:
memory --> up to 6.4GB/sec --> memory controller on CPU
then seperately
bridge chip --> up to 6.4GB/sec --> CPU
So in essence the A64 allows for nearly double the bandwidth of data moving to the CPU. 6.4 GB/sec from the memory and a second 6.4GB/sec from the Hypertransport bus that connects the chipset. On the P4 the same bus must share data path to the CPU with memory / IDE controllers PCI bus, AGP bus, etc...