M3/Pro/Max L2 and SLC sizes?

[Basic75]

Do we have numbers for the L2 and SLC cache sizes of the M3, especially the Pro and Max? At least some of the numbers that are now on Wikipedia must be bogus (32MB for both M3 Pro and Max) since the L2 is per cluster, not per chip or per core. If we run with the new 6-core cluster size I'd expect the M3 Pro to match the M3 at 16MB, if they don't partially disable it on the base M3 and possibly the binned M3 Pro, and only the M3 Max to weigh in at 32MB. But enough speculation, has anybody got reliable numbers?

 

[FunnyElevator]

There are no official numbers, but it’s possible to extrapolate some info by comparing the die shots of the M3 Gen with the previous.

M3
L2 Cache big cores: 16MB
L2 Cache little cores: 4MB
System Level Cache (SLC): 8MB

M3 Pro
L2 Cache big cores: 16MB
L2 Cache little cores: 4MB
System Level Cache (SLC): 12MB

M3 Max
L2 Cache big cores: 32MB
L2 Cache little cores: 4MB
System Level Cache (SLC): 48MB

I’m a bit uncertain of the SLC numbers. But memory cells don’t shrink as much as logic anymore, so I’m fairly confident that we don’t see any massive increases.

Feel free to verify my numbers by comparing the cache areas on these annotated die shots:
https://techhub.social/@fhausler/111340026036295175

 

[esou]

Hi, I tried running some cache latency benchmarks (https://chipsandcheese.com/memory-latency-test and https://github.com/sudarsunkannan/memlatency) on my M3 Max and I extracted the following conclusions:

- Each 6P core cluster has 16MB of L2, but a core from one cluster can access the cache from the other cluster with a latency penalty.
- The SLC is 64MB.