Workshops Archive

Hitting the Memory Wall: Why Cache Miss Tolerance Defines CPU Performance Now


Workshop: International Workshop on RISC-V for HPC (RISCVHPC)

Authors: Mia, Shu-Yu Chang (Andes Technology USA)

Abstract: The memory wall is the defining bottleneck in today's AI, automotive, data center, and HPC SoCs. CPUs operate at GHz frequencies with single-digit nanosecond cycle times, yet DRAM latency remains stubbornly at 60–100 ns, and MMIO accesses across complex Networks-on-Chip can take even longer. No single architectural feature can solve this alone — it demands a multi-dimensional defense. This article shares our experience using Google's FlatBuffers library as a memory subsystem stress test on the Andes 46-series RISC-V core. Our evaluation found outstanding transaction capability delivers up to 39% architectural benefit when memory accesses are independent, but only 6% under pointer-chasing patterns. The 46-series attacks this problem from every angle: outstanding transactions to exploit memory-level parallelism, hardware prefetching to anticipate access patterns, and software prefetch to give programmers direct control.


Back to International Workshop on RISC-V for HPC (RISCVHPC) Archive Listing

Back to Full Workshop Archive Listing