Just throw another few hundred RPM on the fans, problem solved. Wait.. Apple won't do that -- too loud?
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
First M2 benchmarks
- Thread starter EugW
- Start date
- Sort by reaction score
That’s not throttling. That’s cinebench not utilizing cores properly.
We know the answer to this. Max put fans on maximum and the speed didn’t change, because…cinebench is not a good benchmark on apple silicon.
I wasn't referring to the cinebench benchmark itself, I was referring to the fan curve itself that Apple uses, but fair point and absolutely agreed.
Hopefully they fix that. People will need their fans turning ever so slightly.
There's no way you will not have throttling under high load on a passively cooled computer, no way around it. My M1 MBA throttled a *lot*, my iPads throttle, my iPhone throttles a lot, my android tablet throttles.
I'm saying cinebench is inefficient on Apple Silicon because...it is. I didn't think that was a surprising comment. It's been discussed here frequently.
It's also been discussed (on Twitter) by an ex-writer for Anandtech (who now works for Nuvia). Here with measurements showing Cinebench underutilising the available power.
Typically throttling is where the cpu/gpu starts at a certain speed then drops down to a lower speed. Max's video shows the cpu starting a certain speed...then staying there for the remainder of the test, regardless of the fan speed. I can't see how that qualifies as throttling. The fact that you feel it's missing 200mhz is irrelevant to the topic of throttling. Not all apps are equipped to use the total power of the cpu. Cinebench on Apple Silicon is one of those.
I stand corrected. Nonetheless, a testing app should force a CPU to it's max CPU freq'. Yet, we don't see that. Another red flag there.
If these testers had access to Spec cpu 2017, it probably would push the cpu. Alas it costs $1200 I think.
I’ll have to go back and watch again, but I don’t recall Max showing cpu freq for other tests.
I guess someone should run this and asitop for a bit and see if frequency drops, that is if it goes to max frequency.
www.tunabellysoftware.com
Endurance: CPU Stress Test
Endurance: CPU Stress Test is the premier macOS app to test thermal limits, CPU load, battery life along with full logging.
Everyone should stop taking Max Tech seriously, they're clowns. They don't know much of anything, they're just good at making slick looking clickbait videos which always amount to a dumbass shouting enthusiastically about things he's misunderstood.
Second, the real thing I wanted to post about is that this behavior where clocks are usually below max rated speed isn't quite what you think. The maximum rated clock frequency is real, but is only available when a single core is active. If two cores are active, another lower limit becomes the max, and if three there might be another limit, etc.
Drilling down a bit, in M1 performance cores are grouped in clusters of four and the performance state of one cluster doesn't affect any other. If I run a single core load on my M1 Max, I see P cluster 0 running at 3228 MHz (the maximum). If I spin up another thread, it drops to 3132 MHz. If I have three or four threads running, it drops to its lowest limit, 3036 MHz. If I add a fifth thread, P cluster 1 comes online and runs at 3228 MHz while P cluster 0 continues to run at 3036.
This kind of thing is a common pattern in modern high performance chip design. Using the obvious example, it's present to a much greater degree in Intel products. It's not exclusively about thermals either; for M1 I'd guess it's more about electrical power distribution. Each core consumes quite a lot of electrical current at high power states, and that produces I*R voltage droop in the chip's power distribution network. Droop means the power supply voltage observed at gates located in the middle of dense high power areas is depressed quite a bit; if you let this go too far some gates will have supply voltage dip too low for correct operation. Managing voltage droop is important, and one way of doing it is to restrict clocks a bit when multiple copies of a high performance, high power block that are physically laid out next to each other are all active at the same time.
Last edited:
See my reply to @CWallace just above. It is completely normal for M series chips under all-core loads to run their P cores a few hundred MHz below maximum. It's not throttling and is not a new behavior for M2.
Exactly...and this is how the Mba will work also
Because even when the fan is in full speed the clock is not go any higher to 3.49ghz for example
Remember that it depends on how many cores are used. A single core is permitted to hit max clocks on the A15 Bionic for example, while 2 cores active limits the clock speed.I guess someone should run this and asitop for a bit and see if frequency drops, that is if it goes to max frequency.
Endurance: CPU Stress Test
Endurance: CPU Stress Test is the premier macOS app to test thermal limits, CPU load, battery life along with full logging.
The high-efficiency cores maxes out regardless of threads (1 to 4).
The same might apply for the M2 but we will probably have to wait for AnandTech or others to do some deep dive on the implementation of these Avalanche and Blizzard cores.
A part of me feels refreshed that Apple markets the frequencies (with 'turbo boost' included) unlike some other confusing marketing schemes. So perhaps we are just all confused by marketing mumbo-jumbo!
As previously mentioned, Cinebench is not a good testing app for Apple CPUs. Results produced using it should be taken with a grain of salt. It’s a shame SPEC is not more widely accessible.
They don’t? Maybe I’m extremely confused, but I have never seen Apple publish or advertise the frequency of any Apple Silicon products?
Oh contraire, competing CPU manufacturers and their buyers consider it a wonderful and completely valid benchmark, because it shows what they already know, Apple bad!