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

Between 2022 and 2024, Ryzen 7 desktop chips achieved roughly a 50% performance jump: Geekbench 6 single-core scores rose from 2,016 (5800X3D, Zen 3) to 2,969 (9800X3D, Zen 5), and multi-core scores from 11,832 to 18,751. That’s about a 47% single-core and 58% multi-core improvement. Clock speeds contributed only a small part of the gain - max boost rose from 4.5 GHz to 5.2 GHz (≈15%) - so other changes drove the bulk of the increase.

Most of the extra performance came from a ≈50% rise in transistor count (≈11B to 16B), concentrated in beefing up the core. Zen 5 widened and deepened the pipeline: dispatch width increased from 6→8, integer ALUs from 4→6, the reorder buffer from 256→448 entries, L2 per core doubled (512 KB→1 MB) and L1 data cache grew 32 KB→48 KB. SIMD capabilities doubled width-wise (four 256-bit units → four 512-bit units) and loads/stores per cycle scaled to match, enabling much more parallel work per cycle. In short, AMD converted extra transistors into wider execution, larger on-core caches, and greater instruction-level and data parallelism; Zen 6 and a reported 256-core Epyc “Venice” with 1 GB L3 hint at further expansion.