Comparison help M4 Vs quad core i7

[Grumpyman]

Hi. Looking for some help finding a comparison chart type info.
Planning to get a M4 Mac mini with 16mg ram in Nov or whenever they are released.
Currently have a late 2013 MacBook Pro 2.6GHz quad-core Intel Core i7 with 16mg ram.
I guess it will be a noticeable bump in power but is there a chart somewhere which shows the relative performance?

Would keep the old one as it works just fine but can’t upgrade the OS anymore.
Thanks!

 

[SoupyTwist]

I've probably got the wrong MacBook, but EveryMac's comparison tool is handy for comparing Macs:

https://everymac.com/ultimate-mac-c...od2=MacMiniSilicon003&prod3=MacMiniSilicon004

It lets you compare Geekbench scores, which is a benchmark for testing CPU performance. Geekbench's benchmark browser may also have the Macs you're looking to compare:

Mac Benchmarks - Geekbench

Also, if you still want to use your MacBook as a portable Mac to accompany your new Mac mini when it arrives, you may be able to update to a new and more secure version of macOS using OpenCore Legacy Patcher:

OpenCore Legacy Patcher

Experience macOS just like before

dortania.github.io

Mr Macintosh's videos on the subject are super helpful, I've used them to keep my 2012 Mac mini going as a file server, Time Machine etc. for my other Macs.

 

[Grumpyman]

Thanks very much, that’s perfect. Exactly what I was looking for.
Would be handy to keep the old dog running so will check out the videos.

 

[Ben J.]

Would keep the old one as it works just fine but can’t upgrade the OS anymore.

A 2013 MBP would not be upgradable beyond Catalina perhaps? If it works for you and you don't have to upgrade because of some app demanding it, don't worry about it. It will have 99,9% of the functionality and sync with the M4 as if it had a current OS version. I haven't done the OCLP thing myself, but it involves some 'hacks' that I would avoid if I have a choice, personally. In theory, your MBP can last forever.

Those apple silicon laptops are sweet, though. Lots of power, cool, silent, long-lasting batteries.

 

[mr_roboto]

Here's a comparison between your machine and one of the leaked M4 MBP scores:

Mac16,1 vs MacBook Pro (Retina) - Geekbench

M4 Mini isn't expected to score any differently so this should be a reasonably good comparison, unless the M4 MBP data is faked (but it probably isn't).

 

[Grumpyman]

Thanks everyone, really helpful info.

Looks like Big Sur is a modern as i can go for the OS.
Main reason for going for a new Mac is so that I can enjoy the latest OS versions and the security/features that come with it.
The Old MBP still works just the same as the day I bought it some 10+years ago (not something anyone with a PC has ever said i'm sure!).
I use the MBP at a desk 99% of the time so a Mac Mini it is and save a few £s vs a new laptop. The old MBP is sill ok for the occasional trip out.

Those spec comparisons from SoupyTwist & mr_robot are insane. 500-700% improvement I shall enjoy that.
Thanks again for your help. I had no idea how to find that stuff out.

 

[dmccloud]

Hi. Looking for some help finding a comparison chart type info.
Planning to get a M4 Mac mini with 16mg ram in Nov or whenever they are released.
Currently have a late 2013 MacBook Pro 2.6GHz quad-core Intel Core i7 with 16mg ram.
I guess it will be a noticeable bump in power but is there a chart somewhere which shows the relative performance?

Would keep the old one as it works just fine but can’t upgrade the OS anymore.
Thanks!

Since M4 is still only on iPads at the present time, there really are not any true comparisons that can be made between it and any other CPU/SoC at this time. Once the M4 officially hits the desktop/laptop lineup, then it would be possible to draw more accurate comparisons between M4 and the Intel CPU in that 2013 MBP.

With that being said, I did look into the Geekbench scores for the i7-4960HQ (the CPU your Mac is most likely using), the base M1 used in the MacBook Air. and the base M3. Performance-wise, that Intel chip got smoked even by the fanless M1 with only 7 GPU cores:

CPU	Single Core	Multi Core
Intel i7-4960HQ	1117	3676
Apple M1 (8CPU, 7 GPU cores)	2343	8351
Apple M3	3082	11581

While benchmarks such as these can not accurately reflect on the user experience using these CPUs/SoCs, they can illustrate the performance capabilities relative to each other. There are other considerations at play here, including battery life, weight/portability, screen quality, etc. However, those are really matters of personal preference in terms of which factors are important and which are not.

 

[Seoras]

I'm writing this on an 2017 27" iMac with 4.2 GHz Quad-Core Intel Core i7.
I'm an app developer and I have one app that is massive (several storyboards and about 1/4 million lines of code)
Takes about 2mins 30s to compile from clean on my iMac.
On wife's M2 MacBook Air (it has 16G of RAM compared to the iMac with 40G) it takes just over 30s.
That didn't correlate to the Geekbench numbers when I compared them before I did the compile test.
So Geekbench says 3x faster but don't be surprised if what you actually experience is up to 5 or 6 times faster.
Yep, I too am holding off upgrading my work horse for the new Mac Mini next week. I've had to wait too long already...

 

[dogen]

Looks like Big Sur is a modern as i can go for the OS.

You can update to Sequoia with OCLP.

 

[Grumpyman]

You can update to Sequoia with OCLP.

Appreciate the info.....

 

[Ed Hartouni]

I did a CPU comparison a while back for various iMac and a MacBook Intel cpus, basically to see the efficacy of putting in a more capable cpu (i5 to i7). The task I used was heavily computational (as my general use cases are), using the CERN Geant4 monte carlo. I setup their basic example B1 and ran it for 2e6 "events". This is an example of "dumb" parallelism, essentially running n applications and seeing the scaling. Having just received my M4 mac mini I ran the exact simulations.

The results on the Intel cpus revealed that the limitations are largely due to the Apple thermal regulation model. So the scaling was very poor, and thermal throttling dominated the time to complete the n applications independent of the cpu type.

For the M4 mac mini, the results scaled linearly, with an inflection at the point for which the efficiency cores were being used in addition to the performance cores. My analysis of this particular case showed that the performance cores were roughly 2 times as fast as the efficiency cores.

The last point for the M4 mac mini shows the effect of sharing, as there were background processes executing at the same time. But for these tests, there was no departure from a linear scaling performance. I'd have to test with a longer run to see if I could get the fan to spin up, but for practical matters I suspect that the thermals probably won't be an issue for most of my usage.

I also ran Geant4 with a single 2e6 event generation, but using multiple "threads." This was most likely on a performance core (the task manager seems to run computationally intensive applications on the performance cores).

This scaling is not quite linear, total time = 69 s/threads^1.1 + 7.6 s, but I am still mulling over the implications.

using these numbers, I'd expect the M4 max (12 performance + 4 efficiency) to be about 2 times faster than the M4 mini (4 performance + 6 efficiency). I am planning on buying a M4 max Studio when they become available.

 

[mr_roboto]

The results on the Intel cpus revealed that the limitations are largely due to the Apple thermal regulation model. So the scaling was very poor, and dominated the time to complete the n applications independent of the cpu type.

It's not the Apple thermal regulation model, that's just how Intel chips behave under multi-core loads.

You know how you'll see a number like 5.1 GHz for the "max turbo" frequency of an Intel CPU? There's a lot of footnotes on that spec. One of them is that you only get the max turbo frequency so long as most of the CPU cores are idle (and thus able to be powered down). The chip's power/turbo controller has a table of clock speed limits indexed by how many cores are active, so as that number goes up, clock frequency comes down.

This scheme started out fairly simple when Intel first adopted it, but has gotten progressively more complicated and difficult to reason about. Still, it has always resulted in signifcantly sub-linear scaling from 1-core performance, even with programs that exhibit "dumb" parallelism. The only way to avoid it is to get an overclocker 'K' model CPU, which allows overriding these default limitations, and spend a lot of money on a very powerful cooling system.

Even Apple's chips have a similar behavior - you only get the highest possible CPU frequency while there's only one core active in a P cluster. However, because Apple's power per core is so much less than Intel's, Apple gets away with a far milder frequency penalty, and hasn't yet been forced to resort to extremely complex 'turbo' algorithms.

 

[Ed Hartouni]

Even Apple's chips have a similar behavior - you only get the highest possible CPU frequency while there's only one core active in a P cluster. However, because Apple's power per core is so much less than Intel's, Apple gets away with a far milder frequency penalty, and hasn't yet been forced to resort to extremely complex 'turbo' algorithms.

It seems that there is no departure from linear in the M4 mini, even with a rather challenging computational task. So it is hard to see the Apple throttle down.

One thing I haven't done yet is to track the power use under these sorts of loads, which would be a good next test.

 

[mr_roboto]

It seems that there is no departure from linear in the M4 mini, even with a rather challenging computational task. So it is hard to see the Apple throttle down.

One thing I haven't done yet is to track the power use under these sorts of loads, which would be a good next test.

It can be tricky to observe because whenever two cores are active in a P cluster, frequency backs off. In modern systems, true single-thread is rare. Random activity (especially but not limited to anything you do in the user interface) happening on the machine will cause additional threads to run now and then; even if they don't have a lot of work to do they block the cluster from running at full speed.

In theory, on any computer with two P clusters, the second thread could run on the second P cluster and both could run at the highest clock speed. However, Apple's scheduler optimizes for power over performance, and thus schedules any second thread to the same P cluster that's currently running a single-thread compute task. (This optimizes for power because some resources in a cluster are shared between all cores, so a completely idle cluster gets to turn off those shared resources whereas a partially idle cluster has to light them up.)

 

[Ed Hartouni]

It can be tricky to observe because whenever two cores are active in a P cluster, frequency backs off.

There is a lot of detail available from the powermetrics output.

Here is the plot for the thread dependence with the CPU power:

each CPU in the clusters can have a different frequency selected from the tables:

E-Cluster HW Active frequency table: 900, 1080, 1476, 1860, 2172, 2496, 2808, 2892 MHz
P-Cluster HW Active frequency table: 912, 1260, 1512, 1800, 2088, 2352, 2616, 2868, 3096, 3300, 3468, 3624, 3756, 3876, 3936, 3984, 4044, 4104, 4416, 4464 MHz

which are dynamically scheduled. The aggregate for this study:

where the E-cluster (blue points) is used as the thread count goes beyond the number of P-cluster (red points) CPUs.

These runs were too short to turn the fan on (stayed at 1000 rpm, the minimum). I'll run the "dumb parallelism" test and see if the increased thermal load changes the frequency and/or increases the fan speed.

 

[Ed Hartouni]

For the "dumb parallelism" run, I launch N processes at once. This results in N times the events simulated. (The "N thread" test results in the same number of simulated events, but faster).

The results, are essentially the same:

where the one red point is the result of simulating 3 times the number of simulated events for 10 processes. This results in a 3.03 times the other 10 process simulation (the last blue point).

The green points are the associated CPU power values.

The red dashed line is the model that has the P-cluster CPUs 2 times faster than the E-Cluster CPUs; seems to be a good model.

The power usage also seems very similar, with a plateau between 20 W and 25W. The frequency throttling is identical to the other test, here the results are superimposed, the points are on top of each other. (Red for P-cluster, blue for E-Cluster).

The fans did spin up in this test, apparently when the "PSMI Supply AC/DC Supply 1" sensor reaches 102ºC. The fan increases until this temp drops to the threshold. The maximum observed for this test was 2950 rpm (min is 1000, max is 4900).

It would be interesting to increase the computational load to max out the fan and see if there is a larger throttling effect. But this is pretty impressive compared to the old Intel based iMacs. Basically 5 times the computational speed increase for my old iMacs executing this exact same task.

 

[Seoras]

For the "dumb parallelism" run, I launch N processes at once. This results in N times the events simulated. (The "N thread" test results in the same number of simulated events, but faster).
...
<Snip>
...
It would be interesting to increase the computational load to max out the fan and see if there is a larger throttling effect. But this is pretty impressive compared to the old Intel based iMacs. Basically 5 times the computational speed increase for my old iMacs executing this exact same task.

I'm writing this on an 2017 27" iMac with 4.2 GHz Quad-Core Intel Core i7.
I'm an app developer and I have one app that is massive (several storyboards and about 1/4 million lines of code)
Takes about 2mins 30s to compile from clean on my iMac.
On wife's M2 MacBook Air (it has 16G of RAM compared to the iMac with 40G) it takes just over 30s.
That didn't correlate to the Geekbench numbers when I compared them before I did the compile test.
So Geekbench says 3x faster but don't be surprised if what you actually experience is up to 5 or 6 times faster.
Yep, I too am holding off upgrading my work horse for the new Mac Mini next week. I've had to wait too long already...

Got the M4 Pro Mini with 14 Cores. That App now compiles in 25s. So 5 to 6 times faster.
Archiving even quicker which is nice.
Yet to hear the fan other than when I ran Cinebench's GPU tests and even then it was much quieter than the old iMac.
Apple Watch said 50db held right up against it. Other's have reported lower levels so take with a pinch of salt.

@Ed - what's involved in running your tests and producing those graphs?
If not too time consuming I could run them on my Mini.

 

[Ed Hartouni]

@Ed - what's involved in running your tests and producing those graphs?
If not too time consuming I could run them on my Mini.

Thanks Seoras, but there's a lot of stuff to download, etc... and then making sure you have a FORTRAN complier etc... I use these applications for research so I have them available to compare across lots of computers I use.

I'm thinking it isn't too worth while for someone who doesn't use them to actually instantiate them. Better is to find some applications you use that are compute intensive and bench them against your previous performance.

But... Geant4 - getting started

As I mentioned I'm looking forward to the M4 max Mac Studio when it is available, so I decided not to go for the M4 pro Mini, which looks like a very sweet machine.