AMD Ryzen's 50% Speedup in Two Years Came From Wider Cores, Not Faster Clocks

Daniel Lemire pushes back on the common claim that CPU progress has stalled, using AMD’s 3D V-Cache Ryzen line as a counterexample. Comparing Geekbench 6 scores across three 8-core parts — the 5800X3D (Zen 3, 2022), 7800X3D (Zen 4, 2023), and 9800X3D (Zen 5, 2024) — he shows single-core performance climbing from 2,016 to 2,969 and multi-core from 11,832 to 18,751, gains of roughly 47% and 58% in just two years.

Crucially, clock speed explains almost none of this: max boost rose only about 15% (4.5 to 5.2 GHz), far short of the measured improvement. The real drivers are microarchitectural. Across the generations AMD widened dispatch from 6 to 8 instructions per cycle, grew integer ALUs from 4 to 6, doubled L2 cache per core to 1 MB, expanded the reorder buffer from 256 to 448 entries, and enlarged the L1 data cache. Zen 5 also doubled SIMD throughput to full 512-bit units with matching load/store bandwidth.

The takeaway is that meaningful single-thread and throughput gains still come from architecture — more instruction-level parallelism, bigger caches, and wider vector execution — rather than from cranking frequency. For anyone writing performance-sensitive code, it’s a reminder that the headroom now lives in how well software exploits wider, cache-rich cores.