Be careful when upgrading from M1 Pro → M3 Pro
The M3 Pro is awesome but keep in mind it has less performance cores than the M1 Pro and M2 Pro chips—in exchange for more efficiency cores. Most of us already knew this, but if you figured all apps are coded to utilize both efficiency cores and performance cores when doing multicore processing—understandably but unfortunately you would be mistaken. Depending on the core app you use to make a living—upgrading may mean worst performance—yes, even on Apple's own applications.
As you can see here, Apple's own Logic Pro won't touch those six efficiency cores. Thus, for Logic Pro users, the M3 Pro is a bad upgrade path; and even those coming from Intel Macs would be better served buying a used or refurbished M1 Pro Mac (which plays 79 tracks vs M3 Pro's 64 tracks—a significant drop in performance two years later).
Apple sells 25 million Macs per year, so its guaranteed there are a few thousand people buying an M3 Pro that may realize disappointment in their respective core apps. I hope this post can inform a few to not waste their money this year (I'm sure the M4 Pro and M5 Pro will pick up the slack, or just jump to M3 Max—although that's something Mac mini fans can't do).
It's not all bad news. Any app coded to include efficiency cores—also—will beat the M1 Pro / M2 Pro. You just gotta do research on your specific app and not simply rely on Apple marketing vaguery. We need more reviewers to specify app performance and to be evidence based instead of squealing about new colors.
Fortunately this video helped put the M3 Pro in an app-specific context for audio recording. I hope there are similar reviewers doing that in your industry.
The base M1 Pro has the same amount of performance cores as the base M3 Pro (6). M3 Pro has more efficiency cores.
This behavior seems to be specific to Logic Pro though.
Programs can't really steer the OS on how they want their threads to be scheduled. Its up to macOS to decide that, and if a program has too many loaded threads then they will end up scheduled in the E-cores no problems.
I don't know why Apple is messing with Logic in this way, maybe to reduce latency?
Ableton is doing this too.
3 of the 6 apps tested did not utilize efficiency cores. Logic Pro was not the only one.
Apparently that is incorrect. It's app specific and code dependent.
I researched this a bit and you can give guidance to the OS via QoS on tasks that steer the decision of P-core vs E-core.
https://developer.apple.com/documentation/dispatch/dispatchqos
I’m unclear what system calls for this look like though.
Did you choose to use all cores in the settings of Logic? If it set to auto, it will not use all.
I am so frustrated. I had a 16 inch M1 Macbook pro base model. My new M3 Pro 14 inch is so much slower with premiere pro. This is not pleasant.
how come? specific scenarios or in all circumstances?
Lots of people use Ableton for live performances with multiple instruments and ins/outs so I can see why they prioritize extremely low latency, been one of the main selling points of the DAW.
However I expect that this will keep making noise and the software developers will give us the option to fully utilize all cores at the expense of higher battery usage and higher latency.
The fact M1 Pro/Max remains the best bang for the buck option (specially in the used/refurb market) is not breaking news anyway, that machine could end up easily as the most revolutionary laptop of the decade.
M2, M3 and the ones going forward will be very good at doing graphics tasks but I really feel that there’s already diminishing returns CPU wise if apps can’t/don’t know how to properly use the processing power increases.
Another example of how raw numbers increases claimed by Apple don’t matter at all in specific real world usage scenarios.
Is that the video where the guy uses the unbinned M1 and M2 Pro vs the binned M3 Pro even though they retailed for the same price? Cool
Edit:
You can’t just skew test results using a faulty methodology and then say, “oh, it would have been the same with the unbinned m3 pro anyways”
Guy clearly paints a picture of the M3 pro being drastically slower at logic, which is not true