Quartus® Prime Pro Edition User Guide: Platform Designer

ID 683609
Date 9/30/2024
Public
Document Table of Contents

4.5.4.1.2. Unacceptable Latency Increase

Processors are sensitive to high latency read times and typically retrieve data for use in calculations that cannot proceed until the data arrives. Before adding a bridge to the datapath of a processor instruction or data host, determine whether the clock frequency increase justifies the added latency.

A Nios® II processor instruction master has a cache memory with a read latency of four cycles, which is eight sequential words of data return for each read. At 100 MHz, the first read takes 40 ns to complete. Each successive word takes 10 ns so that eight reads complete in 110 ns.

Figure 168. Performance of a Nios® II Processor and Memory Operating at 100 MHz


Adding a clock crossing bridge allows the memory to operate at 125 MHz. However, this increase in frequency is negated by the increase in latency because if the clock crossing bridge adds six clock cycles of latency at 100 MHz, then the memory continues to operate with a read latency of four clock cycles. Consequently, the first read from memory takes 100 ns, and each successive word takes 10 ns because reads arrive at the frequency of the processor, which is 100 MHz. In total, eight reads complete after 170 ns. Although the memory operates at a higher clock frequency, the frequency at which the host operates limits the throughput.

Figure 169. Performance of a Nios® II Processor and Eight Reads with Ten Cycles Latency