M4 Pro Chip Benchmark Results Reveal an Extremely Impressive Performance Feat

[nathansz]

M4 Pro GPU has about 30 percent improvement over M3 Pro in Metal

Metal Benchmarks - Geekbench

browser.geekbench.com

good to see it moving closer to amd

it seems kinda crazy that you still have to run intel/amd to get the best possbile graphics performance in macos

 

[nathansz]

you’re not getting the point.

We’re not talking about next year being better , that is obvious . We’re talking about the value of these Pro/Max devices erratically losing its value because of this.

computers aren't financial investment instruments

why does it matter if the lose value once you are using them?

 

[nathansz]

Luckily for you there's no gpu benchmark out there following geekbench 6's philosophy. That would be benchmarking for everyday use, like 4k playback or website rendering, then concluding that a 5060 is indeed faster "for the average user" because of some new codec or instructions. That's essentially the case here. Read my previous comment about how badly geekbench scales compared to other cpu benchmarks. Macrumors should amend this article to explain this.

geekbench's gpu benchmarks don't follow geekbench's philosophy?

 

[veena3]

Hence, I said this is a bad time to get an M series chip. I don't want to go back to "that point" when people kept getting left out 100% slower just after a year. I'd just wait for things to go to "recent days of x86"?


Not for long.. apparently 14 + 20 M4 Pro is enough to outperforms that a year later. Max chip is just that, nothing.

You want to be on the "top of the top" and you cheap in the same time
Don't be cheap and buy top tier Mac every year, lol

 

[Juraj22]

M4 Pro does have impressive results. But it is not faster than M2 Ultra in all regards.
For developers, code compilation is still significantly faster (28%) on M2 Ultra than on 14 core M4. Even 16core M4 is not that fast. (M2Ultra is 8% faster)

New chips have new vector instructions that make certain workloads much faster than before, but code compilation is not one of those.

M2 Ultra 236 klines/sec
M4 Max 16c 220 klines/sec <- still not M2 Ultra performance, but for laptop, excellent result. Let's see thermals first.
M4 Pro 14c 184 klines/sec <- great

M3 Max 16c 186 klines/sec <- respectable result
M3 Pro 12c 122 klines/sec <-reason many thinks M3 Pro is not good enough comparing to M2 Pro or Max, but battery life was better

M2 Max 12c 129 klines/sec
M2 Pro 12c 128 klines/sec

M1 Max 12c 108 klines/sec
M1 Pro 12c 108 klines/sec
M1 70 klines/sec
Intel i9-9980HK 8c 62 klines/sec <- last 8 core Intel macbook pro from 2019, after that we got M1


Just for fun:
Laptop CPUs:
Intel Ultra7 258V 64 klines/sec <- but battery life is good
Snapdragon X Elite - X1E80100 136 klines/sec <- nice, a bit better than M2 Max
Desctop CPUs:
Intel 14900 210 klines/sec <- also works as space heater, great in winter, and might be broken already.
Ryzen 9950x 222 klines/sec <- good performance, not as good as Intel heating rooms.
Server CPUs:
Ryzen Threadripper PRO 7985WX 64cores 414 klines/sec <-server part
Ryzen Threadripper PRO 7995WX 96cores 416 klines/sec <-server part

So for M3 Max owners, M4 Max does not make much sense. For M1 Max owners, you get twice the performance. This might make sense, if money is not a problem, but M1 Max still works great.

M4 Pro 14core laptop version can have max 48GB RAM, that is kind of unfortunate. 64 would be nice as on mini.
Regardless, M4 Pro 14core variant is very good value for money.

I also wish Apple added 5G modem as an option, maybe once they have their own ready we finally get it.

 

[chucker23n1]

Why is it, that in SINGLE Core the M4 max is faster than M4 Pro?

Is there higher clockspeed?

Yes, it appears the Max runs at a slightly higher clock. This isn't unprecedented; the M2 and M2 Pro ran at 3.5 GHz, whereas the M2 Max ran at 3.7 GHz.

 

[chucker23n1]

I can also say as AAPL geek, you failed to internalize how buying an expensive tool, like a fully maxxed out Mac computer, only to get nerfed by a baseline, rapidly changing Mac a year later, is a good investment? 😊

A computer is not an investment; it's a cost center. It starts losing money the very moment you've purchased it.

However, a computer is a tool, and as long as you can derive at least as much value out of that tool as you put into it, that's worth it. That you can eventually get an even better tool is an opportunity cost.

 

[stiligFox]

Slightly off-topic but M-series related. I seem to remember around the time of the M1 launch (maybe before) people saying that any iPhone app or game should now be compatible with a Mac because of the similar/shared chip architecture. But that never really happened. Any ideas why?

Youre actually correct! Every single iPad and iPhone app can run natively on an Apple Silicon Mac. Apple however did create a toggle for app developers to submit apps and specify that they do not want the to run on Mac - and it seems the vast majority of the biggest apps are set this way, like Netflix for instance. It’s really annoying :l

 

[chucker23n1]

Exactly right.

I was planning on buying the M4 Pro Mac Mini. Even though I don't need that power right now and I would probably need some of it in 2 or 3 years.

Then I thought, why should I spend so much on this device now when it will be weaker than the base level device by the time I actually need the power.

Probably because most of us aren't time travelers and don't need a hypothetical device from the future, but the one we can get today.

 

[bradman83]

Yes, it appears the Max runs at a slightly higher clock. This isn't unprecedented; the M2 and M2 Pro ran at 3.5 GHz, whereas the M2 Max ran at 3.7 GHz.

Geekbench shows M4 Pro and Max both clocking in at 4.5 GHz, give or take a few MHz.. What is interesting is differentiation by models. Mac16,7 has results for both M4 Pro and Max, Mac 16,11 just the Pro. All of the 16,11 single core scores are around 35-3600, all of the 16,7 scores are around 39-4000, including the result for the M4 Pro in a 16,7. This could be a sign 16,11 is the Mini and could potentially be throttling more with the smaller chassis.

 

[dmr727]

Probably because most of us aren't time travelers and don't need a hypothetical device from the future, but the one we can get today.

This thread makes it pretty easy to differentiate those that actually use their computers for a purpose, and those that are more interested in the flex of having the latest and greatest.

 

[Juraj22]

Slightly off-topic but M-series related. I seem to remember around the time of the M1 launch (maybe before) people saying that any iPhone app or game should now be compatible with a Mac because of the similar/shared chip architecture. But that never really happened. Any ideas why?

There are several reasons:

-UI does not scale well to big screen, because design guys did not think about it..many don't even target iPads, so even if we wanted to do it, it does not provide good user experience, so let's not do it.

-phone does have some sensors/HW mac don't. We can't do it, sorry.

-support of new platform is cost for developer, more testing, more customer support and we don't have a budget for it

-business reasons (any other reason considering $$$ and people making it)

-technical reasons: macOS as a development platform is not as good/nice as UIKit on iOS, and SwiftUI on macOS is not as good as on iOS. SwiftUI also does not scale very well to complex user interface, where you have much more UI elements on screen, meaning you see a lot of performance problems. This makes development harder, it takes longer to make, so it cost more, and on platform is much less people than on iOS, so it is not worth it. Many just failed to make it work properly.
And this one is completely Apple's fault. SwiftUI is not the silver bullet many of us (developers) think it would be. It is great on watchOS, good on iOS and tvOS, but not as good on macOS.

 

[chucker23n1]

Geekbench shows M4 Pro and Max both clocking in at 4.5 GHz, give or take a few MHz.

Yep, but that's just Geekbench guessing. They manually add the real clock later on.

What is interesting is differentiation by models. Mac16,7 has results for both M4 Pro and Max, Mac 16,11 just the Pro.

Not sure what you mean. Mac16,7 and 16,11 are the M4 Pro; one M4 Max model is the Mac16,5. (I believe the Mac16,5 and Mac16,7 are MacBook Pros, the Mac16,1 is an iMac, and Mac16,11 is a Mac mini.)

I believe on Apple Silicon each SoC configuration always has its own gestalt (the Macn,n identifiers), though I'm not sure.

All of the 16,11 single core scores are around 35-3600, all of the 16,7 scores are around 39-4000, including the result for the M4 Pro in a 16,7. This could be a sign 16,11 is the Mini and could potentially be throttling more with the smaller chassis.

Yep, looks that way. Which frankly is a bit of a bummer, and might be Apple's slightly sneaky way of trying to upsell people to the M4 Studio once that launches. OTOH, those are still really good scores for a computer.

 

[EugW]

Geekbench shows M4 Pro and Max both clocking in at 4.5 GHz, give or take a few MHz.. What is interesting is differentiation by models. Mac16,7 has results for both M4 Pro and Max, Mac 16,11 just the Pro. All of the 16,11 single core scores are around 35-3600, all of the 16,7 scores are around 39-4000, including the result for the M4 Pro in a 16,7. This could be a sign 16,11 is the Mini and could potentially be throttling more with the smaller chassis.

Well, 16,11 also gets well over 22000 multi-core, so it doesn't seem like it's throttling much. The top 16,11 multi-core scores are within 1-2% of the top 16,7 multi-core score, so essentially they are the same.

 

[Manzanito]

Same reason why Apple keeps adding stuff to their chip with Max Ultra Pro Extreme suffix? People bought them so they can have the fastest Mac ever, for whatever reasons.

Problem is 4090 is NVidia's end game for 2 years now and kept its promise. Apple just kept bashing their last gen Pro Max Extreme chip like a cheap toy. I have no problem with Apple kept pushing the limit, but having an M4 Pro running circles around M3 Max in just a year? Apparently a "Max" level chip means nothing.

You don’t buy that kind of computer because it will be the fastest a couple of years from now, but because it’s the fastest you can buy right now.

Another reason would be to future proof it. I doubt M1 Max owners are complaining about how slow their computers are, because those are crazy fast chips.

 

[satchmo]

Base M4 seems to be a touch faster than a M3 Pro in both single and multi-core.

 

[chucker23n1]

Can anyone say how this GPU in M4 Pro compares to like 4060?

The 4060 is about 50% faster.

 

[EugW]

Base M4 seems to be a touch faster than a M3 Pro in both single and multi-core.

View attachment 2445342View attachment 2445343

A touch faster in multi-core, but way faster in single-core.

 

[ProbablyDylan]

I wonder why Apple doesn't scale their Neural Engine the same way they scale their GPU in Pro/Max chips. They mention how a large shared memory pool would be beneficial for something like a local LLM, but NPU in the Max chip is otherwise identical to the NPU with the base M4.

If local AI will become more important as time goes on, why not make more local AI compute available with the more capable chips?

 

[Tago]

So: Assuming you find the base Mini or the base Mini Pro relevant for your purposes.
What did you have to purchase to get equal performance before M4? And what did you have to pay for it? For good measure, use 24gb for M4...

 

[Tago]

I wonder why Apple doesn't scale their Neural Engine the same way they scale their GPU in Pro/Max chips. They mention how a large shared memory pool would be beneficial for something like a local LLM, but NPU in the Max chip is otherwise identical to the NPU with the base M4.

If local AI will become more important as time goes on, why not make more local AI compute available with the more capable chips?

Speculation, but it probably has the juice required for the timespan Apple works with. My guess it is generous capacity now, and ample in a few years. Believe they just have made sure it is good enough for the period of time it is supported - 6-8 years... As far as I understand they have jacked up the capacity without changing the core numbers though...

 

[EugW]

So: Assuming you find the base Mini or the base Mini Pro relevant for your purposes.
What did you have to purchase to get equal performance before M4? And what did you have to pay for it? For good measure, use 24gb for M4...

For CPU multi-core, base M4 is slightly faster than both M3 Pro and M2 Max.

 

[dgdosen]

Apple should bring back Jony to do something interesting with this power (or some near-future generation)

 

[mrothroc]

Another reason would be to future proof it. I doubt M1 Max owners are complaining about how slow their computers are, because those are crazy fast chips.

I'm in that boat. I have a 2021 MBP with an M1 Max. I do code compilation, run k8s with a full suite of containers for an entire SaaS product, occasionally do local inference on it, etc. It handles all that, no problem.

I used to update my computer every two years because the performance gains made a material difference in my workflow, but once I hit this one I realized I am going to be productive with it for a long, long time.

 

[Populus]

A computer is not an investment; it's a cost center. It starts losing money the very moment you've purchased it.

However, a computer is a tool, and as long as you can derive at least as much value out of that tool as you put into it, that's worth it. That you can eventually get an even better tool is an opportunity cost.

Hey! Maybe you (or someone else) can give me a good advice.

I want to finally upgrade from my 2014 Mac mini, and my main usage for raw CPU performance is for compressing or transcoding video (of around 1.2GB) using Handbrake. Other uses would involve learning video editing but not necessarily 4K. And some occasional gaming if more AAA games land on the Mac. Maybe learning digital music creation with GarageBand, or LogicPro/Cubase/FLStudio eventually.

I admit that I want a Mac that lasts me another 10 years (8 years realistically), and for that, honestly, I’d like to equip it with a good amount of RAM. Especially if local LLMs become more mainstream in a year or two. But M4 Mac Mini with 1TB/32GB is 1.639€ here in Spain. But for just 70€ more I have the binned 12cCPU/16cGPU M4Pro with half the storage and less (but faster) RAM. Binned M4Pro which benchmarks we still don’t know but I assume it to be in the range of 20.000 points.

Honestly I don’t know what to buy. I really want the Pro chip, because I usually buy a new mach every 8 to 10 years, I like to enjoy a good dev for many years instead of throwing away or selling my stuff every 2 years, and I really thing is the most environmentally and economically sustainable way to consume.

But I’m not sure I will be comfortable with 512GB of SSD in 5 or 6 years. Heck, I don’t even know if I’ll be comfortable with half the storage of what I currently have, although I have lots of USB-C external SSDs with good speeds. Is it feasible to work directly on external storage? The Thunderbolt 5 ports are promising and in 5 years or less, Thunderbolt 5 external drives are affordable.

Would you go with the 1TB/32GB M4? Or for virtually the same price a 512GB/24GB M4Pro?

Thank you.