Tootfinder

Opt-in global Mastodon full text search. Join the index!

@fell@ma.fellr.net
2025-07-08 09:46:43

I always wondered why my 3900X with 64 MB of cache performs so much worse than a 5700X3D with 96 MB of cache. It's only ⇓ more cache, after all.
Well, `lstopo` had the answer. The cache is split into 4 chunks (CCXs), so a single thread only really has 16 MB of cache available.
I actually managed to gain a few FPS through isolating a group of cores that share an L3 cache by manually setting CPU affinity.
1/3

Screenshot of the "lstopo" utility. It displays a block diagram of the CPU topology. There are various rectangles of slightly different colour. There are 12 cores with 2 threads each, but the cores are organised in 4 chunks of 3. Each chunk shares 16 MB of L3 cache.