CPU Cache Sharing

The cluster discusses shared L3 cache sizes and architectures in high-core-count CPUs, including comparisons between AMD and Intel chips, NUMA-like behaviors, and performance implications of limited cache per core.

📉 Falling 0.4x Hardware
3,215
Comments
19
Years Active
5
Top Authors
#2332
Topic ID

Activity Over Time

2008
14
2009
28
2010
37
2011
31
2012
57
2013
37
2014
93
2015
84
2016
205
2017
169
2018
246
2019
269
2020
357
2021
341
2022
313
2023
367
2024
302
2025
247
2026
18

Keywords

RAM CPU print.html ARM LLC HBM MOSI MERSI GPU channelregister.co cache cores core memory caches latency cpu amd ring clusters

Sample Comments

d110af5ccf Aug 20, 2022 View on HN

Why can't we have both? L4 cache anyone?

L2 cache isn't shared, at least looking at the die picture.

jacksnipe May 25, 2021 View on HN

Not if those CPUs share one cache!

Aren't AMD parts already (cache-coherent) NUMA?

creshal Sep 10, 2015 View on HN

Don't we already hit that on GPUs? Or do they only have shared caches?

nullc Aug 26, 2025 View on HN

Looks pretty cache starved! One sixth the cache per core of 9684x, a two year old chip.

faragon May 14, 2017 View on HN

Looks like server chips with less L3 cache.

qwertox Nov 28, 2024 View on HN

It's most likely a vCPU. So even the caches are shared.

gameswithgo Apr 14, 2020 View on HN

not per core, but per ccx. a ccx is a group of cores. if you need something that isn't in your ccx's chunk of l3 cache you have to hop through the memory controller and pay a latency penalty.

foota Jun 26, 2020 View on HN

Don't modern processors have more cache available between all cores than available to a single core?