sign up log in
Want to go ad-free? Find out how, here.

A look at the range-topping and very powerful Apple MacBook Pro M4 Max

Technology / opinion
A look at the range-topping and very powerful Apple MacBook Pro M4 Max
Apple MacBook Pro M4 Max; photo: Juha Saarinen
Apple MacBook Pro M4 Max; photo: Juha Saarinen

The end of last year was quite busy with travel and work, so the planned review of the MacBook M4 Max that arrived a bit late thanks to courier delays turned into a "what I did during my holidays" task. Well not quite; I actually tried really hard to swim, go for hikes and chop down vegetation instead of looking at computer screens. Despite that, here it is, a deep dive into the most powerful laptop I've tried out.

If you don't want to go through the techie details below, and you're Mac user who wants top performance in a laptop that will last a long while, the MacBook M4 Max is worth checking out. Budget allowing of course.

OK, the details: my review MacBook M4 Max arrived equipped with a chunky 128 gigabytes (GB) of memory, which is the most it can be specified with. Storage is an ample 4 terabytes, and the M4 Max system on a chip (SoC) has a central processing unit (CPU) with 12 performance cores, and 4 efficiency cores for 16 in total. It runs at 4.49 gigahertz (GHz) and is made by TSMC in Taiwan, with a 3 nanometre process - which is very fine indeed, allowing for more transistors, and higher performance with lowered power consumption.

For the graphics and AI applications (see below), the M4 Max comes with 40 cores. The GPU drives a 16-inch Liquid Retina XDR display with 3456 by 2234 pixel resolution, that comes with a particularly great feature in MacBooks for the first time: nano-texture coating. The nano-texture on the display costs $260, and you should definitely add that to your order if you're looking at (literally) MacBook Pros.

It reduces glare heaps, and the nano-texture display has wide viewing angles as well, making it a very noticeable upgrade.

A 96 Watt charger with MagSafe 3 connector is included, and the M4 Max has three USB-C connector combo ports. They support charging, and Thunderbolt 5 and USB-4 for peripherals with both offering up to 120 gigabits per second data speeds.

The USB-C ports can also output full resolution to external displays 10-bit / 1 billion colours, with DisplayPort 2.1 support.

Up to four external displays are supported on MBP M4 Max: Apple says three can run at 6K resolution at 60 Hz refresh rate over Thunderbolt, and one 4K display at 144 Hz over the HDMI connector on the right hand side of the laptop.

Alternatively, you can run three external displays. Two set at 6K/60 Hz over Thunderbolt, with one external screen at 8K/60 Hz; or one external display at 4K/240 Hz hooked up to the HDMI connector, which provides multi-channel audio output as well.

I'll have to try out a multi-screen set up like that sometime.

And yes, the 3.5 mm headphone jack has been kept, and Apple put an SDXC memory card reader on the MBP M4 Max. 

With all the options above added, the price comes to $11,209 including GST. That is of course quite a bit of money, but if you head over to a PC vendor's site and start ticking boxes for a similarly configured Windows workstation laptop, you're looking at a similar expense.

Being Apple, the MacBook M4 Max is a beautifully assembled, premium quality laptop that is built with recycled materials. "No established final assembly sites generate waste sent to landfill as part of Apple's Zero Waste Programme," the company claims.

The 16-inch form factor is one that I like, as it good to work on with a nice screen size, yet not gigantic and difficult to lug around, being fairly slim and weighing 1.62 kilo. It's probably as big as I'd go though, and you can have any colour you want, as long as it's Space Black.

You get great performance for your money from the MacBook M4 Max:

Apple MacBook Pro M3 Max vs M4 Max
Benchmark M3 Max M4 Max Difference (%)
Blender 4.3.0      
CPU  350.84 491.2 +39.92%
GPU  3368.51 5130.78 +52.37%
Geekbench 6      
Single core 3189 3694 +15.85%
Multicore 19361 25474 +31.57%
GPU Metal 125770 190879 +51.75%
Geekbench AI      
CPU Single precision 4449 5971 +34.18%
CPU Half precision 7362 9165 +24.47%
CPU Quantised 6366 7628 +19.83%
GPU Single precision 15347 21929 +42.92%
GPU Half precision 19358 24085 +24.42%
GPU Quantised 14971 18674 +24.76%
NE Single precision 4427 5960 +34.68%
NE Half precision 27027 31208 +15.47%
NE Quantised 29338 38134 +29.97%
ATTO      
Write speed 7.2 GBytes/s 10.6 GBps +47.22%
Read speed 5.6 GBps 7 GBps +25.00%

This is not quite an apples to apples comparison, as the 2023 M3 Max that I tested earlier is a smaller 14-inch model with "only" 96 GB of memory, and with 14 CPU cores and a 30-core GPU. The M3 Max is really fast though, and you can see from the numbers though that the M4 Max is quicker on a single-core basis. 

Memory speed in the M3 Max model I had was 307.2 GBps whereas the M4 Max tested hits 546 GBps using uprated low power double data rate memory on the highly integrated system on a chip. For context, that's extremely fast. It'll be interesting to see how much Apple's M4 Ultra chip, which is said to be due out in June this year, will be. The older M1 and M2 Ultra chips managed 800 GBps memory speeds. 

Apple claims in its marketing that the base model M4 is 50 per cent quicker than the M2 which came out in 2022, and has four times the graphics performance.

Why is the Wi-Fi like that?

All the hardware in the MBP M4 Max is the latest and greatest, except for the Wi-Fi wireless networking. For reasons only Apple knows, it's not the newer Wi-Fi 7 standard, but the earlier Wi-Fi 6E. The latter is plenty fast with support for 160 MHz wide channels in the 6 GHz band, but not as quick as Wi-Fi 7 that can use up to 320 MHz worth of bandwidth.

Wi-Fi 7 also adds other features like lowered latency; Apple's iPhone 16s support the newer Wi-Fi standard, although they're limited to 160 MHz channels.

What can you do with an MBP M4 Max?

That's easy: you can enjoy any amount of comments about how nobody needs that much computing power and/or memory, and how this-and-that computer from 2015 still works fine and does everything needed. It is compulsory for most computer users to state such things, no matter their level of geekiness.

On a more serious note, if you earn your crust as a developer, video editor or similar, the sizeable speed up the MBP M4 Max provides over older gear is great for professionals. 

There's more than enough performance to run multiple virtual machines, and the graphics unit has support for hardware ray tracing and mesh shading. What that means is that instead of creating hyper realistic scenes in software or getting the CPU to do the work, it's offloaded to the fast GPU which has special circuitry for the purpose. The Blender 3D software benchmark score is among the top seven per cent the open source organisation has recorded.

Of course, the game (or simulator) has to know the hardware ray tracing and mesh shading features are available so that it can use them, to get the performance boost. Ditto the hardware AOMedia AV1 decoding support which, without going into too much nerdy detail, is a newish, state-of-the-art video compression and transmission format.

If you do audio/video work, the MBP M4 Max has a really good six speaker sound system with "force cancelling woofers" for bass, wide stereo, spatial audio and a three microphone array with directional beam forming. And, Dolby Atmos sound processing support. It sounds great.

A new use case has appeared over the last few years, which is running artificial intelligence large language models (LLMs) locally for testing and development, or because you can. The MBP M4 Max runs biggish models like DeepSeek-R1 and Meta AI's Llama 3.3 with 70 billion parameters fine thanks to the 128 GB of unified memory, and fast storage speed. If you work with AI models, you'll appreciate what it means, being able to run big ones on a laptop.

However, models of that size hit the GPU cores hard, as I noticed in the past. DeepSeek-R1 in particular gets the fans in the MBP M4 Max going as it reasons, trying to figure out if you're asking the model something politically sensitive that it should decline to answer.

Why isn't the uprated Apple Silicon Neural Engine in the M4 Max, which can do 38 trillion operations per second (TOPS) used for LLMs? ASNE is aimed at accelerating smaller AI tasks like image and speech recognition, and Apple Intelligence of course, while being economical with power.

The benchmark table above shows that the GPU is good at single-precision floating point operations, while the ASNE is quicker with half precision and quantised operations. Yes, AI gets really complex like that, and there are other factors as well to consider as you try to run big models that are 40+ gigabytes in size on your own machine.

Long story short, everything happens near instantly on the MBP M4 Max, with little or no delay. Interestingly (well, for me at least) is that running day-to-day stuff for work, the Activity Monitor tool indicates around 55-65 GB of memory is in use by macOS and apps (mostly web browsers), with very little data paged out to disk. 

Apple uses the same sized 72.4 Watt-hour battery across the M4 MacBook Pro line; with the M4 Max, you get an estimated 13 to 18 hours worth of battery life depending on use, as opposed to 14-22 hours with the M4 Pro, or 16-24 hours in the M4 models that contain less powerful and complex chips.

In practice, I was able to make the M4 Max last a working day, although I wouldn't run a big LLM while on battery. Some trade-off there in other words, but considering the high specification and powerful chip, it's not much.

The MacBook Pro M4 Max lives up to its name, providing an impressive platform for professionals in a compact and portable package. It should've had Wi-Fi 7 though, but that's the only minor grizzle.

We welcome your comments below. If you are not already registered, please register to comment.

Remember we welcome robust, respectful and insightful debate. We don't welcome abusive or defamatory comments and will de-register those repeatedly making such comments. Our current comment policy is here.

11 Comments

I am amazed how good the 1.5B Deep Seek is locally, on a older M1 2020 Macbook.

SW dev on an M4 with 70B helping via Roo Code would be worth looking at. 

 

Up
3

I've seen the same about smaller models that have been "distilled" to perform better. Roo Code looks interesting. 

Up
0

Not many people have $11k to spend on a laptop. Even less need to.

Up
3

Agree, but If I am paying a developer 150k and they can double there output its a cheap investment.  I could reduce a team of 10 to 7 and still have more productivity, AI Generated documentation etc etc

This is threatening to young java coders, but may be a godsend in the Cobol space, where its getting harder and harder to find new coders, many of the 65 yo coders want to keep working and they are incredibly productive now, this type of enhancement is worth paying for.  

There is also no requirement for the model to always run on the local laptop, it can be shared in cloud as well.

Up
6

I see VMs touted - but aside from being able run more, is it actually any better than the M1/M2?

I had an M1, then M2 Pro for work, and regularly used VMs for product testing. Whilst useful to have, they were not fast - especially when compared to the KVM box running on 10yo ex-leased hardware in the corner. In fact, main reason for having the Pro was multiple monitor support, not available on the cheaper model.

Still, wouldn't say no if someone gave me one of these... 😀

Up
1

Unfortunately the performance numbers may end up reducing a bit after the SLAP and FLOP vulnerabilities that were reported to Apple months back but haven't been fixed/mitigated yet. Similar to the Spectre vulnerability, mitigations could reduce performance significantly.

See https://predictors.fail/

Up
0

I saw about SLAP and FLOP. Side channel attacks are a pain, and it'll be interesting to see if Apple can come up with a mitigation that doesn't tank performance.

Up
0

$11209?

If you actually need the power this laptop gives, that enables processor-heavy functions like video processing, mathematical modelling and things like FEA (finite element analysis) on mechanical designs, why wouldn't a cheaper, equally capable desktop be more appropriate? You're going to want to plug it in to big high-res screen and a beast of a storage system that will be fixed objects.

The only application I can think of is if you need to run things like ERP system server emulations at client premises.

Or is it really a massively overpowered gaming laptop for the well-heeled?

Up
2

I registered Saarinen Web Services already, being entrepreneurial and all that. But anyway, Apple's point is that you can take that computing power with you, like when you need to do video edits etc out in the field. If that's not a requirement, then you could do it another way. I've seen people cluster Mac Minis with M3/M4, to run big, big LLMs for example. 

Up
0

This is the tech equivalent of a gatcha game whale

Up
1

Still no numeric keypad? .... Still off my shopping list.

Up
0