Achieve 10X DRAM Bandwidth Improvement with a DDR Controller Read Reorder Buffer
Achieve 10X DRAM Bandwidth Improvement with a DDR Controller Read Reorder Buffer
The read reorder buffer (RRB) is a silicon-proven architectural enhancement available in Synopsys uMCTL and uMCTL2 DDR memory controller IP products. This white paper will explain the concept of the read reorder buffer and explain how a read reorder buffer can improve memory bandwidth. It then concludes with experimental results showing how DRAM controllers with different architectures can achieve vastly different DRAM bus utilizations of 10%, 66%, or 100% from the same input traffic stream, depending on whether the architecture has no RRB, an RRB with external scheduling, or an RRB with content-addressable memory (CAM)-based scheduling.
Table of Contents
- Introduction
- Reordering transaction in a DRAM controller
- Restriction on transaction reordering
- Write reordering and read/write reordering
- Experimental setup
- Experimental results
- Potential concerns
- Conclusion
Please complete the following form then click 'submit' to complete the download.
Note: all fields with * are required