|ARM Technical Support Knowledge Articles|
Applies to: ARM9 processors
Yes, it makes a significant difference to performance.
ARM9 processors generally achieve around 1.1 DMIPS/MHz when executing Dhrystone code entirely within their Level 1 memory system, which has a Harvard architecture.
However, even with zero wait-state memory, running from the bus incurs cycles of latency in the bus interface and contention between instruction and data accesses, either in the bus interface where the Level 2 interface is a single shared (Von Neumann) interface, or outside of the processor where the bus has a shared fabric or where the instruction and data accesses compete for access to the same memory slave.
For example, ARM968E-S will run Dhrystone approximately 80% faster from Level 1 Tightly Couple Memory (TCM) than from zero wait-state memory on the AHB bus running at processor speed.
Did you find this article helpful? Yes No
How can we improve this article?