"The most surprising part of this design is that the
SPE's 256K local store is not a cache -- it is actually the full
amount of memory that an SPE has to work with at a time for both
programs and data. This seems like a disadvantage, but it actually
gives several advantages:
"Local store memory accesses are extremely fast compared to main
memory accesses.
Accesses to local store memory can be predicted down to the clock
cycle.
Moving data in and out of main memory can be requested
asynchronously and predicted ahead of time"