proomt

Search

Search posts, papers, and topics

All posts

Hacker News front pageView all posts by Daniel Lemire2 min readintermediate

How did AMD Ryzen get 50% faster in two years?

Read on lemire.meHN 247·91 comments

Summary

AMD’s Ryzen 7 desktop CPUs (Zen 3 → Zen 5) gained ~50 % performance in two years by widening the core (dispatch width 6→8, integer ALUs 4→6, reorder buffer 256→448), expanding caches (L2 512 KB→1 MB, L1 data 32 KB→48 KB), and doubling SIMD width (four 256‑bit units → four 512‑bit units). Clock speeds rose only 15 % (4.5 GHz→5.2 GHz). Transistor count grew ~50 % (≈11 B→16 B), mostly in the cores.

  • Performance gains came from architectural widening rather than higher clock speeds.
  • Larger L2/L1 caches and a bigger reorder buffer improve instruction-level parallelism.
  • Doubling SIMD width and load/store bandwidth directly benefits data‑parallel workloads.
  • AMD’s roadmap hints at even larger core counts and cache sizes in future Zen 6 designs.

Understanding where performance improvements come from helps engineers make realistic expectations for future CPU upgrades, informs software optimization strategies (e.g., leveraging wider SIMD), and provides context for hardware‑aware performance engineering.

6/10
architectureperformancecpuamdzen

Related reading

  1. SiliconBench: Speed, Memory, and Fidelity for LLM Serving on Unified-Memory Desktops

    SiliconBench benchmarks nine Apple‑Silicon LLM serving engines on unified‑memory desktops, measuring speed, memory usage, and output fidelity across Qwen3, Qwen3.5, and Gemma‑4 models. It finds vllm‑metal doubles throughput at modest concurrency, memory budgets often fail to preserve headroom, and only three stacks satisfy all fidelity and model‑coverage requirements, with tensor‑parallel scaling…

    Hugging Face Daily Papersarxiv.org1 minpaper

  2. Benchmarking Wild vs Mold

    Reproduces Mold’s linker benchmarks on a 16‑core Ryzen, shows that configuration (filesystem, delete‑output, fork) explains most of the Wild vs Mold speed gap, and notes recent Mold releases and upcoming Wild tweaks that close the gap.

    Lobstersgithub.io4 minHN461lobste.rs52

  3. Accurate Models of AMD Matrix Cores

    The authors reverse‑engineer AMD’s CDNA matrix cores, build per‑architecture MATLAB models that match hardware bit‑for‑bit on 10 M random tests, and use them to compare AMD vs NVIDIA tensor‑core accuracy.

    Hacker News front pagearxiv.org2 minpaperHN7911

  5. What Every Programmer Should Know About Memory

    The paper explains how modern CPU caches, memory controllers, and NUMA architectures affect program performance and what developers can do to write cache‑friendly code. It provides concrete advice and tooling to reduce cache misses, avoid false sharing, and place memory near the accessing cores.

    Hall of Famefreebsd.org397 minpaper