Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Apple M3 Pro Chip Has 25% Less Memory Bandwidth Than M1/M2 Pro
- Thread starter MacRumors
- Start date
- Sort by reaction score
I like that Apple is comparing to the M1 because basically nobody who bought the right M2 machine for their needs should be replacing it with an M3 machine at this point. If you missed and bought the wrong M2 machine (or your needs have changed) then go get a replacement and take the financial hit. But your miss was probably in areas like lack of RAM or you bought too low a spec when you got the M2 machine. So it isn't just the CPU improvement that is driving your upgrade decision.
And this is scary!
First they pushed people to upgrade the SSD or not choosing the base model due to slower SSD performance. Now they are pushing people to get the Max. So M3 has better performance than M1 and M2 only if we go for the Max version?
I ordered a M3 Pro 11-core. The M1 works fine for me in the Mac Mini and for a laptop I think the M1 Pro strikes the best balance between power and efficiency, especially now that it provides 6 efficiency cores. Four performance cores would be enough for me so one is a bonus. I will not notice the lower memory bandwidth for sure. I stuck with 18GB RAM and hope it will still last me 7 years like my 15'' MacBook Pro 2016 I7/16GB did.
The M1 and M2 only supported Thunderbolt 3 and the Pro/Max chips support Thunderbolt 4. It's the same with the M3 - the base chip supports Thunderbolt 3, the Pro/Max supports Thunderbolt 4.
To be labeled Thunderbolt 4 the device must be able to support dual 4K displays via a single port. Since the base chip is limited to one additional display Apple can't call it Thunderbolt 4.
It’s more like they aren’t able to make it better than it is. TSMC 3nm is too expensive and the yield is so bad that the a17 pro is a joke and that lead directly to the m3 being cut down a lot.
It's nice that Apple offers laptop choices, at various performance levels and price points.
Simply purchase what you actually need.
Need more performance and features? ---> Pay more money.
Don't need the best performance/features? ---> Save some money.
Easy.
Not a difficult concept to understand.
Simply purchase what you actually need.
Need more performance and features? ---> Pay more money.
Don't need the best performance/features? ---> Save some money.
Easy.
Not a difficult concept to understand.
Last edited:
Air of desperation? I think it is Apple giving the single finger salute to all the pundits that claim they know Apple, that trade in rumors based on history rather than facts. It is Apple trying something new, and you guys think it is 'rushed.' lol. People complain about the first day online ordering log jam, so now Apple has it so you can order immediately. No log jam. And they are trying something shorter, to the point, without a huge production of it all. Maybe more events in the year? They certainly caught everyone by surprise. Just how is it they finalized the chip design for three classes, M3, M3 Pro, and M3 Max, and got them into the MBP and iMac and NO ONE PREDICTED IT? That's not rushing.
Credit where credit is due, Apple is shaking things up. Not rushing. I can't wait to see what's next. They certainly got a lot of free advertising this new way.
Why would someone with a two year old high-end laptop feel the need to upgrade already? Computers in general should have a lifespan of 4-6 years at a bare minimum before they need to be upgraded unless you really didn't spec out your needs properly when you purchased. These are targeted at people with the later model Intel Macs. M1 owners shouldn't need to upgrade until at least the M4/M5 series and by then they'll see a sizable performance boost. Hopefully this is a sign that Apple plans to roll out new M chips on a more frequent basis so these incremental updates pile up.
Same - picked up a refurbed MPB14 M1 Max 64GB 2TB a bit ago to replace my still-working-well-but-hot MBP16 Intel. I like the space black as well, but not paying $$$$$ for a color. I have little doubt it'll out-perform my M1 Max, but that's ok... in general my M1 Max performs 'more or less' like an M2 Pro (but with more RAM) - I'm quite OK skipping this generation and seeing what the M4 Max brings in the future.I bought MBP because I want to use multiple external screens but not editing vids. Didn't need M2 so picked up a greatly discounted M1 Max config with 32G/1TB which is plenty for me and should last for some more years. The "space black" colour looks nicebut M3 Max not worth it to upgrade...for me. Plenty of ppl who will be happy with it though, especially if coming from Intel Mac. Just the pure silence, even under load is a bliss.
I'll probably recommend 14" with M3 Pro base though instead of 15" Air. 15" Air not cheap cheap to upgrade to similar ram/hdd.
Pretty much, but for a reasonably large number of not-so-much-to-moderately-computer-literate, the higher the number, the better. Realistically, it is all a bit odd though regardless, including the Neural Engine 'downgrades' and lack of direct comparisons to M2 series.
I do think for whatever reason they has some 3nm process issues to sort through but were wanting to get <something significant (price-wise)> out the door in time for the holiday ordering frenzies.
I like the space black, but I skin the topside/display and my keyboard/trackpad portion of the case anyways, then tier leave it or peel it depending on skin condition if/when I eventually sell it down the line, so - yeah, don't care that much, but I'd probably pick it whenever I 'need'/want to replace my M1 Max in the future, and not seeing any real reason to do so with the M3.
I was looking at the features of the various models and Apple seems to go out of it's way to not say "Thunderbolt 3", instead it says "Thunderbolt / USB 4". Is that meant to confuse the buyer into thinking it's Thunderbolt 4 and USB 4? Sneaky of Apple. But like you said, both are the same speed.
Then you're making very bad choices on what you buy. My Dell desktops last 3-5 years easily and I usually replace them while they're still working. I can add memory, upgrade storage, upgrade graphics, put in better WIFI card, etc. You can't change a thing on a Mac. What you buy on day 1 is what you keep for 5 years. And you get to pay more for that privilege.
It’s likely that memory decompression and/or increased cache performance ameliorates this. Memory busses take up a lot of space and not an insignificant amount of power.
There was a giant leap from intel -> M1, at least in most configurations and I think people have unrealistic expectations for the jump between each generation ever since. These chips, including the M3, are still super powerful in particular with their world beating performance per watt for laptops. I think the M1->M3 is a nice jump for 3 years and is likely the earliest anyone would update their computer anyway.
Well...part of the original 'bang' from the M1 besides being ARM based, it the die process (5nm vs Intel 11 or 14mm at the time IIRC). Reducing the process size will always reduce power consumption and give a bit of performance boost for the same chip, so the combination of both ARM-based (already low power consumption by design/history) and the smaller process made for a compelling story alone.
The jump from Apple 5nm to 3nm undoubtedly netted some power savings and increased by itself, but it does seem the chips architectural evolution hasn't also made significant improvements at this point, just more incremental. I'm waiting for some third party comprehensive benchmarks, but I don't see much compelling to me as an M1 Max owner right now. Of course, when I need to replace my wife's Intel MBA soon-ish, I'll include the M3 in the 'maybe' list, but I suspect I'll wind up with an M2 or even M1 refurb for the performance per $.
Like Intel used to have their Tick-Tock pattern (significant arch updates, followed by die shrink and incremental improvements.., replaced by kind of tick-tock-tock with the last being incremental improvements and optimizations), it seems Apple is trying to work out their own pattern. Having said that - yeah, am curious to see what the M4 will actually bring - minor/incremental changes or larger scale improvements/changes.
Not sure they intentionally 'gimped' the M2 Pro, but overall when even the Neural Engine nodes and others are incremental or borderline backpedaling, I do think they hit a time crunch to get something out for holiday sales. Entirely possible some of it is down to the transition to TSMC's 3nm process as well.
And of course Apple will benefit in reduced costs by obsoleting the 13" "pro" by no longer needing to produce either the Touch Bar or the 13" MBP chassis. The 'not far from next model up $' has kind of always been there, followed by the shock of RAM and storage upgrade costs.
Oh you'd be surprised how many people will go through each iteration, just because they can.
I bought an M1 MBA on launch day with some discount, sold it for a very good price when it was just over two years old, so it still had some Applecare left on it and replaced it with an M2 MBA. I won't be getting an M3 though.
Unironically: if the efficiency cores are good enough to make this work this is actually the ideal, no split of core types, just a super efficient cpu
The memory bandwidth is probably not going to make very much of a difference for most real world workflows. It's actually quite difficult to saturate the full memory bandwidth of these chips, especially at 400GBPS.
I remember seeing a lot of independent benchmarks and tests done on the M1 Max that couldn't saturate it even with fully maxed out CPU and GPU workloads. (Theoretically, maxing both at the same time would probably push it to its limit on certain kinds of memory-heavy workloads, though I never saw a benchmark that actually attempted this. Someone has probably attempted it somewhere since then, it'd be interesting to see.)
I'd be curious to see how the 150GB/sec ones perform though. Typically, CPU workloads rarely use this kind of bandwidth, but intensive GPU workloads can in certain circumstances. Then again, the M3 Pro doesn't have as many GPU cores as its much beefier M3 Max counterpart, so it'll be interesting to see the benchmarks to see how the M3 Pro performs with the reduced bandwidth.
I remember seeing a lot of independent benchmarks and tests done on the M1 Max that couldn't saturate it even with fully maxed out CPU and GPU workloads. (Theoretically, maxing both at the same time would probably push it to its limit on certain kinds of memory-heavy workloads, though I never saw a benchmark that actually attempted this. Someone has probably attempted it somewhere since then, it'd be interesting to see.)
I'd be curious to see how the 150GB/sec ones perform though. Typically, CPU workloads rarely use this kind of bandwidth, but intensive GPU workloads can in certain circumstances. Then again, the M3 Pro doesn't have as many GPU cores as its much beefier M3 Max counterpart, so it'll be interesting to see the benchmarks to see how the M3 Pro performs with the reduced bandwidth.