Inside M4 chips: P cores | Hacker News Reader

Inside M4 chips: P cores

eclecticlight.co

35 pointsingve9 comments

From the article:

Key information

    Current M4 chips offer 4-12 CPU P cores.

    M4 P cores are arranged in clusters of up to 6, sharing L2 cache and running at a common frequency.

    P core clusters can be shut down, idling at their minimum frequency of 1,260 MHz, or at one of 18 set frequencies up to a maximum of 4,512 MHz, as controlled by macOS.

    Their instruction set is “technically” ARMv9.2-A, but without its Scalable Vector Extension (SVE).

    They use 1-2 mW when idle, rising to peaks of 1,400 mW (floating point) or 3,230 mW (NEON vector code).

    macOS preferentially allocates them threads at all QoS higher than 9 (Background).

    Threads running on M4 P cores are mobile, and may be moved to another core in the same cluster frequently, and after just over a second may be transferred to a core in the other P cluster, when available.

    Thread mobility makes interpretation of the CPU History window in Activity Monitor very difficult.

I don't understand the thread mobility. I only see the overhead but surely there must be massive gains in performance by spreading the thermal load.

If a thread is blocked waiting for something, then at some point in the future it will be ready to continue execution. But what if some other thread is still running on that specific core? If there is some other core available and you can restart execution there with low overhead, that's a performance win.

I think that as asynchronous code becomes more and more common, this kind of capability becomes more and more beneficial.

Yes, I assume that it is better to spread the generated heat across the die. Migrating threads between clusters probably does have some cost (mostly cache misses, minimal cost to pausing and moving) but if done less than every second that is likely miniscule.

Given their experience in designing custom CPUs it seems like a lost opportunity for Apple to not invest in building its own cloud (like AWS)? Perhaps offer cost effective compute using these designs? Does iCloud running in Apple datacenters or do they use other clouds?

Private Cloud Compute is essentially just that, their own silicon, but for AI-only use cases and only for Apple.

https://security.apple.com/blog/private-cloud-compute/

iCloud uses public clouds like AWS and GCP. Apple Intelligence Private Compute (or whatever it’s called) uses Apple’s own servers running Apple Silicon. It will be interesting to see if they expand that offering over time.

afaik Apple is on google's cloud. some guy told me that google engineers refer to Apple as "bigfoot" internally, due to its size as a customer.

I believe they are using m2 ultra studios to provide cloud based AI services.

Reply on news.ycombinator.com