How is this considered throttling when everything else shows it matching or beating the Pro, including in the extended version of the same test? Admittedly I know nothing about these tests and whether they're more CPU or GPU heavy or equally heavy, but that seems like a strange conclusion
M5 Pro and Max unveiled
- Thread starter leman
- Start date
Jimmyjames
Elite Member
- Joined
- Jul 13, 2022
- Posts
- 1,491
What’s strange is the standard test show the 14” take a beating, but the extended one is more equal. I don’t know what to make of it.
Edit I see you answered this above!
Edit 2 hmmm still confused.
Agreed! The only other thing I can think of is IF it's thermals and he ran the tests consecutively that the machine was overheating then cooled before the next test. But we're entering the realm of pure baseless speculation on my part.
And I also don't know enough about these tests. PugetBench is not something I've ever really looked at too much.
The only thing similar is that CB R24 result where the M5 Pro drew more power than the M5 Max and got an equal (slightly worse) score. Though that was the 16" Pro misbehaving in my opinion, not the 14" Max (which again the 14" result makes sense for throttling - the Max used less power than it should've and got a worse score than we've seen the Max get elsewhere).
Last edited:
I'm gonna be direct I've completely ignored notebookcheck this entire time. Everything I've said is either based on a multitude of other reviews or your own posts of their tests.
I hope you've seen my comments regarding Logic Pro. It is the most likely explanation for what's going on beyond the 14" chassis typically having less thermal headroom since it's smaller than the 16", I'd be curious to know what you think about that
exoticspice1
Site Champ
- Joined
- Jul 19, 2022
- Posts
- 457
Logic only uses 12 cores. 6 super and 6 performance cores. The other 6 are idle.
My point was that before Logic Pro couldn't use any HE cores no matter how many there were. Now it can. What changed? I'm legitimately asking
exoticspice1
Site Champ
- Joined
- Jul 19, 2022
- Posts
- 457
I don’t know. There were DAWs that could use the E cores before like Reaper and now Reaper can use all the 18 cores in the M5 Pro unlike Logic.
But also previously there where apps could use all cores that were available and max them out but now are not able to do so with M5 Pro due to a cpu firmware issue.
WRT software optimization being needed? Possible, hard to tease out and it wouldn't explain some of these results - *some* of the 14" results especially have all the hallmarks of how processor behaves under throttling. It's also possible that multiple things are going on. Honestly, this feels very much iPhone 15 (I think it was the 15) which suffered initial teething issues and Apple fixed after a few months - i.e. if memory serves Apple screwed up the power curves so the phone was getting to hot and then performance suffered and had to be reduced further than it would've been if it hadn't gotten so hot. It's also possible that the computer is overcompensating too quickly - i.e. thinking it's about to overheat and reducing performance, but it's actually fine. There are too many possible explanations.
I don't know about Reaper. But something had to have changed between M4 and M5 that Logic Pro can now use HE cores. Logic Pro cannot utilize E cores on M5, only P and S cores, as far as I can tell from peoples tests. But I guarantee this is also an app issue. As I was going to write to @dada_dave @Jimmyjames
I've just seen a new YouTube video going over Adobe software, specifically LR classic, and someone commented they noticed not even on M4 Max that adobe utilizes all of their cores. So if there is a firmware issue, it's also being affected by app optimization too
exoticspice1
Site Champ
- Joined
- Jul 19, 2022
- Posts
- 457
Yeah could be both. But I’m hopeful Apple will resolve it in a couple of weeks by working with app devs
Regardless if there is an issue or not with Apple's firmware, or if you got a bad chip, or whatever else, if you don't feel happy with a product,
you deserve to be happy. So I am sorry it didn't work for your needs.
Did you submit your complaint to AppleCare and/or the Apple Store when returning?
exoticspice1
Site Champ
- Joined
- Jul 19, 2022
- Posts
- 457
No I didn’t but I think I’ll write to Craig. Probably will reach the executive liaison team quicker, it worked well in the past for me. They respond quick.
Here's a hypothesis: Traditionally Apple had one E cluster and it was always mostly there to do low priority background work. These chips have two M clusters and zero E clusters. Perhaps they've chosen to designate one M cluster as the substitute for the old E cluster, meaning it's the one that all the low-priority background work gets scheduled to. They're probably trying to keep it from clocking up unless the scheduler runs out of cores on the other two clusters and starts scheduling higher prio work on the pseudo-E cluster, at which point the limits get taken off.
Assuming this speculation is true, I could see bugs in either or both of applications and the scheduler resulting in failure to utilize more than 6 M cores for highly multithreaded loads under some circumstances.
I've just seen M5 Pro obliterating Intel 358H with B390 in a Samsung laptop by 2-4X faster performance across the gamut of tests, running 1 celsius degree cooler, not spinning fans up at all or not often like Samsung, running 20% brighter (600 vs 500), and still having the same battery percentage (100 wH vs 80 wH) at the end.
It's amazing
It's amazing
I've also seen the M5 Max with 40 cores obliterate the M3 with 80 cores in diffusion models, on battery no less:
Photos:
20 billion diffusion model 16 bit with 4 steps, LoRA:
M5 (40 cores): 11 seconds
M3 (80 cores): 19 seconds
20 billion diffusion model 16 bit with 30 steps, LoRA:
M5 (40 cores): 122 secs
M3 (80 cores): 206 secs
Videos:
22 billion diffusion model:
M5 (40 cores): 121 secs
M3 (80 cores): 206 secs
Photos:
20 billion diffusion model 16 bit with 4 steps, LoRA:
M5 (40 cores): 11 seconds
M3 (80 cores): 19 seconds
20 billion diffusion model 16 bit with 30 steps, LoRA:
M5 (40 cores): 122 secs
M3 (80 cores): 206 secs
Videos:
22 billion diffusion model:
M5 (40 cores): 121 secs
M3 (80 cores): 206 secs
Nycturne
Elite Member
- Joined
- Nov 12, 2021
- Posts
- 1,772
My guess is that Logic did something stupid when it counted how many threads to spin up, and only counted the performance cores. Possibly by looking for how many cores were called "Performance" or something similarly goofy. If accurate, it would also explain this comment:
Count the cores, see 12 performance cores, spin up 12 threads. Whomp whomp.
- Joined
- Sep 26, 2021
- Posts
- 8,180
- Main Camera
- Sony
My understanding is, as far as the operating system is concerned, the “S” cores are named “P,” and the “P” cores are named “E.”
I don't think the internal software refers to "performance" or "super" but probably just specific codes, but regardless I don't think that would explain it only using half of the Performance cores, because it would use all of them then. It users 6 super and 6 performance
I'm assuming Logic Pro was updated to account for the new stuff.
NotEntirelyConfused
Power User
- Joined
- May 15, 2024
- Posts
- 208
I thought the P cores were named "M"?
