7 Comments

285k doubled p-core core-to-core latency when comparing to 14900k, which also has 12 L3 slices.

285k L3 latency is similar to 14900k, but loses to 14900k over a variety of worklords.

It'd be great if you could further examine it when intel rolls out all 'fixes' and PMU events are published.

Expand full comment

I second that! Intel has mentioned ("leaked") that their upcoming firmware update should fix some of the almost embarrassing lack of performance uplift vs. Raptor Lake. If that firmware update could actually reduce cache latencies, it'd go some way to achieve this.

Expand full comment

So ever since Intel moved to non-inclusive cache design, their strategy is to emphasize the private caches at the expense of the L3, while AMD's huge and fast L3 cache and relatively modest private caches strategy seems to gain more net performance.

Expand full comment

I love this article.

Expand full comment

And this completely OT: anyone here have any information on the specs of the N150 (the successor of the Alder Lake N100). One can already find some mini-PCs with it advertised, but nothing yet on Intel Ark (?!) . I know, not the hottest or fastest CPU out there, but these small all-E-Core CPUs/SoCs are basically "it" for small x64 systems.

Expand full comment

First, thanks for this analysis! Second, I'd like to learn more about the influence or aor lack thereof of the "SoC" tile on RAM access speeds.

And, OT: if anyone here has a source (die shot or link to one) on the relative size of a Lion Cove P-Core vs. a Skymont core or four core cluster, please post it - thanks!

Expand full comment

Bravo! Impressive.

Expand full comment