Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

_alxmac

macrumors newbie
Original poster
Sep 24, 2021
15
12
I've been trying to find an answer online but I am having a hard time understanding what Memory Bandwidth actually does.

I'm getting the laptop for music production so I don't really need the 24GPU cores that the cheapest Max delivers but there's also the memory bandwidth difference. What does a 200GB/s difference mean in overall performance and will I benefit in any way of having bought the M1 Max even though I only work with audio? I bought the Max just to be sure because I don't want to miss out on any performance when I'm already spending so much money on a machine. I bought the M1 Max 24 core, 32GB Ram 2TB SSD.
 
Well, for one thing, it could "help" the CPU to work faster when there is a memory demanding application.
Think of it this way:
CPU = factory
slower memory bandwidth = ocean vessel
faster memory bandwidth = plane
memory = customer

Factory can work as fast as it could but it won't help reach the customer since it could be blocked by slower ocean vessel.

This does not mean that you'll always feel the performance gain. From some of the tests I'd seen, it was more notable for media creation application and games, but with single percentage performance gain (read: no bottlenecking by the memory bandwidth).
 
The M1 Max has 4 chips for a 512-bit memory bus and the M1 Pro has 2 chips for a 256-bit memory bus.
 
This is an area of interest to me As I have ordered to 16” MacBook Pro with 32 GB ram. For an extra 200 I could have had the Pro Max chip. My use case is coding and potentially running VMs and Docker instances. I have no need to connect more than two external monitors.

Watching a couple of the techie threads on here I have seen it suggested (more than once) that the standard CPU cores may not be ”wide enough” to utilise the increased memory bus the Pro Max provides. In other words, this increased transfer speed may just be a benefit to the GPU cores.

The above assertion is consistent with Apple’s text on the sales page on the hyper link “which chip is right for you”. Ie they talk about graphic intensive workflows.

Hopefully the benchmarks next week will provide further insight. If I have made a mistake I will simply return my order within the 14 days and reorder .
 
  • Like
Reactions: ascender
I came across this today.

For CPU workloads the M1 already has sufficient bandwidth. Each CPU core runs instructions sequentially, and it's hard to predict for sure what data they'll need next so they're generally more sensitive to latency than bandwidth. By comparison, GPUs are wildly parallel and often more predictable in their memory access patterns. They're good at hiding latency, but can consume a ton a bandwidth.

In addition the SSD in these machines is delivering data at a little over 7GB/s, so the Pro's memory bandwidth can easily keep up with that while keeping the CPU fed and still leave the majority of bandwidth available for the GPU.
 
Anandtech published the below article which suggests the M1 max has 4 SLC (System Level cache blocks). For reference the M1 Pro has two.

Will be interesting to see the benchmarks come in and see what if any gains are to be had for non GPU related tasks.

https://www.anandtech.com/show/1701...m1-max-giant-new-socs-with-allout-performance

M1MAX_575px.jpg
 
It looks like Geekbench shows the M1 Max w/ its 400GB/sec has been getting about 10% better Multi-Core scores than the M1 Pro w/ its 200GB/sec.

I guess we'll see better numbers when there are more benchmarks that test memory-intensive applications and anything GPU-intensive.
 
It looks like Geekbench shows the M1 Max w/ its 400GB/sec has been getting about 10% better Multi-Core scores than the M1 Pro w/ its 200GB/sec.

I guess we'll see better numbers when there are more benchmarks that test memory-intensive applications and anything GPU-intensive.
10%? Looks like it’s almost a wash, maybe about a 1% difference.
 
It looks like Geekbench shows the M1 Max w/ its 400GB/sec has been getting about 10% better Multi-Core scores than the M1 Pro w/ its 200GB/sec.

I guess we'll see better numbers when there are more benchmarks that test memory-intensive applications and anything GPU-intensive.
Where are you seeing this? They look to be almost the same.
 
I'm not 100% sure what I am interrupting bench mark wise, but to a lay person as I am; I'm not seeing any notable difference between the Pro and Pro Max with regards to the Single and Multi Core numbers.

It will indeed be interesting to see whether the additional "System Level Cache" offers any enhanced performance with regards to the 'standard' cores.

A nice Thunderbolt dock might be possible for the £200, meaning I have a tidier desk.

Apple sure know how to pitch their price points to induce endless FOMO. :)
 
  • Like
Reactions: mctrials23
Its definitely one of those things where you think "...only £200? Sure I might never ever see the benefit, but for £200? Bargain!".

They certainly do. I nearly cancelled my order and placed it again with the Max chip because "its only £200 on top of £2800" but it would have knocked back shipping by another week and I am currently using a temporary mac mini that I need to return to apple in the next 2 weeks.
 
There's a good comparison tool half way down here


9798C1CE-BA05-469D-AE87-87920187FF8D.png 97161EF4-E830-4090-80F3-57EB8E16C863.png

You can cycle through various pro software to get an idea of improvements on offer in CPU and GPU. Looks clear that unless you're working on something GPU intensive there's little benefit of max over pro, CPU performance is identical across both in the graphics. Yeah it's marketing and not proper benchmarks etc, but if there was a notable difference you'd expect Apple to highlight it.
 
I guess it’s a good move to pre order this now through apple’s edu page for automatic discount and wait till December for M1 max 64gb? I’ve been looking at bestbuy constantly and pickup option never becomes available for m1 max 16” 32gb. Was gonna wait till release day if stores will have some or store pickup would be available but I guess they’ll be oos and forever be shipping only until next year? I’ve never really bought a MacBook before on Black Friday but I’m guessing they don’t really do discounts, instead do bundles or free gift card? I heard 64 ram could help with windows emulation for games so I’ll go for 64
 
I guess it’s a good move to pre order this now through apple’s edu page for automatic discount and wait till December for M1 max 64gb? I’ve been looking at bestbuy constantly and pickup option never becomes available for m1 max 16” 32gb. Was gonna wait till release day if stores will have some or store pickup would be available but I guess they’ll be oos and forever be shipping only until next year? I’ve never really bought a MacBook before on Black Friday but I’m guessing they don’t really do discounts, instead do bundles or free gift card? I heard 64 ram could help with windows emulation for games so I’ll go for 64

If you’re going to playing a lot of games via Windows, I’d strongly suggest against this device, it does not sound like the device for you. If you really want an Apple device, instead of getting a 64GB maxed out MBP, I’d suggest to get a good gaming laptop for the games and an M1 MacBook Air for other things.
 
  • Like
Reactions: osterix
1. I've been trying to find an answer online but I am having a hard time understanding what Memory Bandwidth actually does.

2. I'm getting the laptop for music production so I don't really need the 24GPU cores that the cheapest Max delivers but there's also the memory bandwidth difference. What does a 200GB/s difference mean in overall performance and will I benefit in any way of having bought the M1 Max even though I only work with audio?...

Just re-iterating the questions, is I wonder also to just go with the Pro.

Q1. What does Memory Bandwidth actually do in plain words ?
Q2. Does it benefit an music producer or musician to have the Max ?

The answers I found were these:

A1 It has a larger capacity to move around large data quickly.
A2 This guy has chosen to go with the Max, because of the memory bandwidth. I think he is right, if this is your goal: To maybe push your instrument libraries to the max, in my opinion. Hear him here:
 
Last edited:
On paper, the 400 GB/sec looks like a big deal. Double the memory bandwidth! Wow! But in terms of CPU, the Pro and Max are benchmarking identically, even with contrived, artificial benchmarks, so clearly 200 GB/sec is more than enough.

The 32 GPU cores of the Max are a different story, and the 400 GB/sec is likely needed to support the high memory needs of that set-up.
 
  • Love
Reactions: norwaypianoman
I'd like to read about the subject deeper, cause I'm also confused and this arises me questions regarding GPU performance, from what I've seen, M1 Pro 16GPU in Metal pure benchmarks is just some points down compared to an RTX 3070 Mobile, which effectively has same 256-bit bus as the mentioned SoC, but if you look at the numbers, RTX 3070 obtains nearly 450GB/sec on bandwidth, of course it's GDDR6 but I'm kinda shocked.

I think I can understand of how non-SoC architecture works and drives me a little nuts, cause even just PCI-Express 5.0 can drive just till 128 GB/sec, if you add that to RAM bandwidth (DDR5) + CPU memory controller bandwidth (12900K for example), it can get to barely 200-220ish at max, and this is of course, it doesn't performs as an exact sum but I hope you get my point.

My doubts are regarding M1 Pro 16GPU vs M1 Max 24Cores, is that extra bandwith gonna help in non-intensive GPU memory uses? I mean, when playing video-games at non 4K resolution like 1080p or 1200p, do you think just that extra bandwidth (not counting the cores) is really gonna help moving more frames? It's an honest question, cause I can't understand that Metal benchmarks 100% given the specs, so any information is welcome.

edit: spelling
 
Last edited:
  • Love
Reactions: norwaypianoman
My doubts are regarding M1 Pro 16GPU vs M1 Max 24Cores, is that extra bandwith gonna help in non-intensive GPU memory uses? I mean, when playing video-games at non 4K resolution like 1080p or 1200p, do you think just that extra bandwith (not counting the cores) is really gonna help moving more frames? It's an honest question, cause I can't understand that Metal benchmarks 100% given the specs, so any information is welcome.
The extra bandwidth is only likely to help under extreme loads. Andandtech's tests show they had trouble getting the machine to use more than 200GB/s, but they did manage it.

As for whether the 24-core is worth it in the 16", it also has two video encoders instead of one, which can help in some heavy video work. For gaming, the extra cores will matter, not the extra bandwidth. Don't know if I've seen any direct comparisons of the 24-core 16" and the 32-core 16", though.
 
The extra bandwidth is only likely to help under extreme loads. Andandtech's tests show they had trouble getting the machine to use more than 200GB/s, but they did manage it.
Yeah, I saw that, they had to push really hard the CPU + GPU to exceed those numbers.
As for whether the 24-core is worth it in the 16", it also has two video encoders instead of one, which can help in some heavy video work. For gaming, the extra cores will matter, not the extra bandwidth. Don't know if I've seen any direct comparisons of the 24-core 16" and the 32-core 16", though.
I know regarding amount of encoders and also expanded cache, my question was more towards the 14". I had an internal debate when exchanging M1 Pro 14" 10CPU/16GPU/16GB/1TB for 32GB RAM version, and I almost got trapped to get M1 Max 24GPU (I've read about the throttling on the 32GPU done on 14") just for casual gaming on low resolutions applying the same recipe of "well, it's just 250EUR more ..." but I didn't want to sacrifice batt life or thermals. So I started a search. CPU and memory bandwidth ofc kept me really satisfied coming from DDR4 world. Regarding battery, there's a lot of speculation rather than the two main youtube vids testing this specific, but from users, you can see 5-15% on normal use, being a 10% drop average on M1 Max 24/32 vs M1 Pro 16GPU, and with intensive use, the things get worst. Thermals were also in the range of 5-10ºC hotter on some intensive tasks, so for me was a no/no, more on the thermals side (i'm kinda obsessed with them) rather than battery. M1 Max users, please correct me if I'm being wrong here.

Since I work as a software engineer, and my specifics doesn't involve any video/3D/ML stuff, my only concern was regarding getting some more FPS very very casually, and after some researches saw the RTX 3070 comparison (just on Metal of course) and I felt pretty satisfied and decided to stick my M1 Pro order, since my "hardcore gaming" comes on a Mac Desktop with a GTX 1080, and not on 4K neither, but on 1080p, I pretended to emulate some of that on my new MBP once in a while. After doing some math, I guess I covered my spectactions and not to feel buyer's remorse and I won't need more "bandwidth"

But when looking at the specific bandwidth the RTX3070 Mobile could do, really impresses me and makes me think if they continue to improve, will I be covered with a TOTAL amount of 200 GB/sec with this M1 Pro? Not planning to exchange TBH (it was a pain to wait close to 3 weeks after returning btw :D) just engineer curiosity, and since hardware is not my main field, I tend to be cautious, so if anyone can drop even more information to my doubts, I'd feel really happy, cause maybe I'm really lost about my "math".

PS: New one arrived yesterday, so still testing but so far runs as good as the exchanged one.
 
  • Love
Reactions: norwaypianoman
Regarding battery, there's a lot of speculation rather than the two main youtube vids testing this specific, but from users, you can see 5-15% on normal use, being a 10% drop average on M1 Max 24/32 vs M1 Pro 16GPU, and with intensive use, the things get worst. Thermals were also in the range of 5-10ºC hotter on some intensive tasks, so for me was a no/no, more on the thermals side (i'm kinda obsessed with them) rather than battery.
Things only get worse if the Max is doing more work than the Pro. For comparisons where both are doing the same task at the same level of performance, the difference in battery consumption remains smallish. Same should apply with the thermals, since those relate to the amount of work done. May actually be cooler when operating at the same performance, since the Max has a larger heat sink.

For the 14", the Max 24/32 still performs better for graphics-intensive work than the Pro, as expected, but not as well as the 16" Max versions. Either by design or natural effect of more limited power components, the graphics cores run slower under high load on the 14" Max compared to the 16".
 
Things only get worse if the Max is doing more work than the Pro. For comparisons where both are doing the same task at the same level of performance, the difference in battery consumption remains smallish. Same should apply with the thermals, since those relate to the amount of work done. May actually be cooler when operating at the same performance, since the Max has a larger heat sink.
Yeah, I've also seen that, with a light light use, diffs are more in the margin 3-5% rather than 15% for the most extreme case I've read, that's why I set a baseline of 10% as an average gain for the Pro with a "normal" use (not saying all the time will be that light) Of course, M1 Max will drain battery faster but also will complete tasks quicker, so one hand washes the other. As I said, not needing GPU rather than the use mentioned, so every gain in this aspect given the fact I wanted 14" for me was very important, at the end, if you're gonna need the M1 Max power, that should be the less of the concerns.

Regarding the thermals, it is true what you say about the heatsink, despite that, since I'm talking purely on 14", the package is smaller and hasn't that much space/mass as 16", checked that with a light use, diffs are negligible or non existing, but when using CPU/GPU mid/intense, M1 Max on the 14" has to spin the fans faster to keep it on-the-line, and even that I saw some spike differences 6-8ºC higher (tests seen always in the 14"), and more fan activity also affects battery life.

TBH with you, my personal feel is M1 Pro "belongs" to 14" and M1 Max suits better on the 16" given the improved airflow and a bigger battery if you really need the Max power, but of course, some folks need all that juice in a compact form factor so some lil sacrifices should be made, but that's just my opinion.

For the 14", the Max 24/32 still performs better for graphics-intensive work than the Pro, as expected, but not as well as the 16" Max versions. Either by design or natural effect of more limited power components, the graphics cores run slower under high load on the 14" Max compared to the 16".
With the information we have now, it seems clear that they did in fact put a limit on the M1 Max 32GPU on the 14" version, and maybe if you decide this form factor, the sweet spot clearly goes to 24GPU, improved bandwidth as you mention for intense memory graphic usage, and more GPU cores that can drive more performance to software that uses that graphic power.

Anyway, I'm still interested in the bandwidth stuff from a math/physics point of view as stated before from a pure FPS prospective, I'll keep reading and I find something interesting, share it with you guys.
 
Last edited:
With the information we have now, it seems clear that they did in fact put a limit on the M1 Max 32GPU on the 14" version, and maybe if you decide this form factor, the sweet spot clearly goes to 24GPU, improved bandwidth as you mention for intense memory graphic usage, and more GPU cores that can drive more performance to software that uses that graphic power.
Probably the most 14" performance for the price, at least. The 32-core 14" will still offer more performance, but often not proportionately more.
 
Probably the most 14" performance for the price, at least. The 32-core 14" will still offer more performance, but often not proportionately more.
Yup, I know, specially when you jump in the 32GB boat as I had to do, Apple pricing this time, was a piece of art, they knew what they're accomplishing IMHO, and I think if I weren't this obsessive particularly with temperatures (I'm a crazy watercooling addict on desktops since 13 years ago, so make your guess) and/or gaining even 5% more batt life in the worst case scenario, the price diff was so negligible that it's impossible that your brain doesn't act as "why not? even I don't profit the max bandwidth/extra gpu is quad-channel vs dual-channel" and jump in.

But thinking clearly, for my uses, CPU wise are mostly the same in "numbers", and with 200GB/sec and dual-channel for my kind of work focused on full-stack web development, and also an amateur musician/photographer, I'm more than overpowered for the next few years, and given the fact for what I'd wanted the GPU power (gaming maybe once-twice a month on a low res) ... I'd rather spend it on the AppleCare+ which is clearly maybe more valuable to me, or save it for a Mac Mini/iMac version with even a more overpowered M1 Max chip, and really play there, that would be sweet as rumors quote out.

Just out of curiosity, what's your opinion on my case with all facts exposed? Would you stick with the 14" M1 Pro 10CPU/16GPU or swap it to M1 Max 10CPU/24GPU? As I told you, I made my mind clear, but eager to hear experiences/criticisms

PS: sorry for the off-topic OP and others
 
Last edited:
  • Love
Reactions: norwaypianoman
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.