Apple M4 chips

[David (PassMark)]

My M4 Max R23 single core result is 2303

Which is slower than an average 285K.

M4 Max even worse then M3 Max

They are roughly the same within the margin of error. Might explain why Apple compared it to the M1 when they did the M4 release.
Likely there is a thermal limit or RAM bandwidth bottleneck. Apple has put the priority on making the computer cases smaller and smaller (or thinner). This makes cooling much harder, which in turn leads to thermal throttling. Literal form over function.

USB transfer speeds from a SSD aren't normally CPU limited. So it isn't a good CPU benchmark.

I predict they M4 will look better when they put it into a Mac Pro full size case with proper cooling.

[andrewelick]

My M4 Max R23 single core result is 2303.
In your Single Thread Performance chart M4 Max even worse then M3 Max that simply can't be...

Cinebench 2024 single thread score:
285k - 150
M4 Max - 182

Blender compressed big file open time from same ssd attached to same port speed:
285k - 1 minute 3 seconds
M4 Max - 0 minute 47 seconds

I can continue if you need

[David (PassMark)]

They are close in performance. Meaning it is easy to cherry pick a couple of results and make a claim that one or the other is better. Different algorithms will prefer one CPU or the other.

To illustrate the point, the 285K is better in Cinebench R23 ST than M4 Max. (2209 vs 2347). But Cinebench is s rubbish benchmark, it is far too narrow in it's instruction set. But it happens to match our result, so I will cherry pick it.



Also close enough that using slightly faster or slower RAM could switch the rankings. Or using different BIOS versions or settings. Or different cooling solution.

[andrewelick]

Looks like we're stuck here.
OK. Just one more question.
Why in all other single thread benchmarks M4 Max is faster then 285k, but in your benchmark it slower? And in all real apps also, I was mention before, not just benchmarks.
If you don't believe me, Google it or do this tests yourself as I did.
In my last conclusion, there is definitely something wrong with your single thread ARM benchmark. It's far far away from real world performance results. Thanks.

[David (PassMark)]

Our Benchmark details can be found here.

We can't comment on your code. But I do know that C# and .NET aren't common programming platforms for MacOS / M4 hardware. We are using XCode 12.5, which is more common.

but looks like it simply don't like ARM processors

We rate the M3/M4 are the fastest laptop class CPUs ever made for single threaded apps. Not sure how you can claim we don't like them.
And they are also only 5% below the 285K. So easy to believe some algorithms show more than a 5% benefit on one or the other platform. Especially if your specific 285K isn't configured in an optimal fashion.

Even Apple doesn't believe it is the fastest (for everything all the time). You can be sure they would have published the data if it they thought it was.

[andrewelick]

I'm talking workloads of single core CPU performance only. In all my tests M4 Max faster then 285K by about 20%.
I did some my own simple benchmark in C# for fun. Compiled it for x64 and ARM under .NET 6. Same result. M4 Max faster about the same 20% as other my own programs I use. Converters, compression and other algorithms.
I don't know your single core performance algorithm, but looks like it simply don't like ARM processors.
Again, very strange that your benchmark show different results.

[David (PassMark)]

Blender & Cinema 4D aren't single threaded software. There is also a good chance they are using the GPU as well. So you can't use any claim about Blender to prove single threaded superiority.

And if you look at the actual numbers for Blender, then the dual Epyc's & Xeons completely destroy the M4 (192-Cores for the AMD EPYC 9965). So the M4 isn't the best for threaded apps either. In fact even the M2 Ultra is better than the M4 in Blender.

You might have drunk too much Apple Cool Aid. Even Apple had to compare it to the M1 to make it look good.

I could totally believe there are some apps where the M4 is faster than the 285K however (depending on RAM speeds, GPU in use, number of threads & if the test runs long enough for cooling to be important). M4 is a good CPU. But fastest in all software, no. Not even close.

Yes, PerformanceTest for Mac is ARM native (and Apple has the fastest laptop chips in our benchmark).

[andrewelick]

I must disagree.
If talking single core performance M4 Max definitely faster then 285K.
I got M4 Max and 285K. M4 Max is faster everywhere.
In all other benchmarks and most importantly in all production software I use also, like Blender, Cinema 4D and my own custom software.
It about 20% faster single-core everywhere in comparison to 285K.
Only your benchmark show different results.
Is your MacOS client compiled natively for ARM?
Strange.

[David (PassMark)]

They aren't at the top because they are not the fastest.

I know Apple claimed their CPU is the "industry best single threaded", but they only tested, in their words, "competitive systems and select industry-standard benchmarks".

So their claim was that they are better than some other CPU, which they didn't, name on some benchmark they also didn't name.

Then they claimed the M4 was "1.7x faster than before for daily productivity". But only compared to the M1. Why did they compare the M4 to the M1, instead of to the previous M3 generation?

In our opinion this is deliberately misleading.

A more accurate claim might be, fastest low power Laptop class CPU in single threading. But I guess that doesn't sound as good. Except it seems that some of the M3 chips are actually faster than the M4. So even making this claim probably isn't accurate.

So maybe the actual accurate claim is, "fastest low power Laptop class CPU in single threading except for our older M3 chips". But this really doesn't sound good for marketing. So I can see why they didn't go with the something accurate.

To be fair, they are still great chips. But their marketing department got carried away.

[andrewelick]

Hello. They should be at the top in single-threaded rating. Why they don't? Thanks.