There is a performance overhead to the RSP interface. It involves adding the JTAG module, with a thread sensitive to each JTAG clock and the GDB server module to process all the packets.

Even with minimal debugging activity (just loading the program and running to completion), the fastest the Ackermann's function model would execute was 93 kHz. The same optimized model was used without the debug interface in Embecosm Application Note 6 and achieved 130 kHz.

This is an inescapable overhead. However it is independent of the size of the underlying model. It shows significantly with the OpenRISC 1000, because of the efficiency of the underlying processor model. With larger and more complex SoCs it would be a less significant overhead.