The M1's memory is LPDDR4X-4266 or LPDDR5-5500 (depending on the model, I guess?) which is about double the frequency of the memory in the Intel Macs.
Apparently, this alone seems to account for a lot of the M1's perf wins — see e.g. the explanation under "Geekbench, Single-Core" here: https://www.cpu-monkey.com/en/cpu-apple_m1-1804
Bleeding-edge-clocked DRAM is a lot more costly per GB to produce than middle-of-the-pack-fast DRAM. (Which is weird, given that process shrinks should make things cheaper; but there's a DRAM cartel, so maybe they've been lazy about process shrinks.)
Apparently DRAM and NAND do not shrink as well because in addition to transistors in both cases you need to store some kind of charge in a way that is measurable later on - and the less material present, the less charge you are able to store, and the harder it is to measure.
That's a high frequency, but having two LPDDR chips means at most you have 64 bits being transmitted at a time, right? Intel macs (at least the one I checked), along with most x86 laptops and desktops, transfer 128 bits at a time.
> Apparently, this alone seems to account for a lot of the M1's perf wins — see e.g. the explanation under "Geekbench, Single-Core" here
That's a vague and general statement that site always says, so I wouldn't put much stock into it.
Am I missing something?
Mind you with 16Gb, Docker won’t be that useful.
Then again I would have expected them to have discussed it as much as the video editing.
I am guessing that they’d need a M2 type chipset for accessing more RAM for that. Or maybe they’ve got a new way to do virtualisation since that is such a key thing these days.
Edit: thanks for pointing that out though, that’s why I mentioned it
https://developer.apple.com/documentation/virtualization
https://news.ycombinator.com/item?id=23922846
And the mentioned Virtio here:
http://www.linux-kvm.org/page/Virtio
How well this fits in with current virtualisation would be interesting to find out; I guess this will be for a later version of Big Sur, with a new beefier M2 chip.
On iOS they can get away with less RAM than the rest of the market by killing apps, relaunching them fast, and having severely restricted background processes. On Mac they won't have that luxury. At least they have fast SSDs to help with big pagefiles.
With the heterogeneous memory, your 8GB computer doesn't even have its whole 8GB of main system memory.
When the touchbar MBP launched in 2016 people were already complaining that it couldn't spec up to 32GB like the competition. Four years later, and it's still capped at 16GB.
Hopefully they can grow this for next year's models.
The difference was incredible
Have you looked at Activity Monitor to see what is butchering your memory?!
Second: web browsers, which can easily grab 5-10GB by themselves or even more if RAM is available.
So in other words: everything.
For example Chrome will scale the amount of RAM it reserves based on how much you have available.
Cache is excluded from just about any tool that shows RAM use, at least on desktops. If the ram shows as in use, the default assumption should be that it's in active use and/or wasted, not cache/preloading.
> For example Chrome will scale the amount of RAM it reserves based on how much you have available.
Which features are you thinking about that reserve ram, specifically? The only thing I can think of offhand that looks at your system memory is tab killing, and that feature is very bad at letting go of memory until it's already causing problems.
I'm not a mac user but that seems ridiculous. I'd be investigating what's hogging it all.
I have chrome, Firefox, photoshop, vs code, docker and a few other things running. As a kid I had to manage RAM. As an adult, I buy enough RAM to not need to think about it.
I was committed to buying an M1 on day one. I won’t buy a machine with only 16gb of RAM.
I'm betting this is due to Thunderbolt controller and PCIe lane capacity. They couldn't do four Thunderbolt ports with the M1 SoC, so they dropped the ports. Having four USB-C ports but only two supporting Thunderbolt would be a more obvious step back from the previous MacBook Pro. This way people can just blame it on Apple doing Apple things, instead of seeing a technical limitation.
Yes, it seems crazy, yes it's a lot of ram, but I like to be able to run VMs locally and not have to boot up instances on AWS (insert provider of choice), I like to keep tabs open in my browsers, I like not to have to close apps when I'm using them and I like my computer to be snappy. 64 GB allows that 16 doesn't, 32 barely does.
I thought with the last update they'd finally seen the light and moved to 512/1tb, now we're back with the silly 256gb.
If you factor in having to upgrade ram to 16gb and ssd to 512 it's only £100 shy of the old price. Good, but not as good as it looked to begin with.
Shocked they're still selling the two port machine, it's been nothing but hassle for me as someone who has to use one.
I'm hoping it can wait for v2 of the MacBook Pro 16"
Apple's second version of everything is always worth the wait.
And they will have significantly upgraded CPU/GPUs to match the memory.
Think web developers, photographers, bloggers etc.
And while I understand that many people are stuck on photoshop, I bet it would be easy to beat 800MB by a whole lot. But so I can grasp the situation better, how many non-adjustment layers do those professional photographer use? And of those layers, how many have pixel data that covers more than 10% of the image?
For a model photo shoot retouch, you'd usually have copy layers with fine skin details (to be overlaid on top) and below that you have layers with more rough skin texture which you blur.
Also, quite a lot of them have rim lighting pointed on by using a copy of the image with remapped colors.
Then there's fake bokeh, local glow for warmth, liquify, etc.
So I would assume that the final file has 10 layers, all of which are roughly 8000x6000px, stored in RGB as float (cause you need negative values) and blended together with alpha masks. And I'd estimate that the average layer affects 80%+ of all pixels. So you effectively need to keep all of that in memory, because once you modify one of the lower layers (e.g. blur a wrinkle out of the skin) you'll need all the higher layers for compositing the final visible pixel value.
Still, an 8k by 6k layer with 16 bit floats (which are plenty), stored in full, is less than 400MB. You can fit at least eleven into 4GB of memory.
I'll easily believe that those huge amounts of RAM make things go more smoothly, but it's probably more of a "photoshop doesn't try very hard to optimize memory use" problem than something inherent to photo editing.
Also, your "could be stored in a compact way" is meaningless. Unless your name is Richard and you've designed middle out compression, we are where we are as end users. I'd be happy if someone with your genius insights into editing of photo/video data would go to work for Adobe and revolutionize the way computers handle all of that data. Clearly, they have been at this too long and cannot learn a new trick. Better yet, form your own startup and compete directly with the behemoth that Adobe is and unburden all of us that are suffering life with monthly rental software with underspec'd hardware. Please, we're begging.
> Also, your "could be stored in a compact way" is meaningless. [...]
That's getting way too personal. What the heck?
I'm not suggesting anything complex, either. If someone copies a layer 5 times and applies a low-cpu-cost filter to each copy, you don't have to store the result, just the original data and the filter parameters. You might be able to get something like this already, but it doesn't happen automatically. There are valid tradeoffs in simplicity vs. speed vs. memory.
"Could be done differently" is not me insulting everyone that doesn't do it that way!
It was surprising to see essentially the same form factor, the same operating system and not much to distinguish the three machines presented (lots of repetition like "faster compiles with XCode").
BTW, what's the size and weight of the new Air compared to the MacBook (which I liked, but which was killed before I could get one)?
Seeing two machines that are nearly identical reminds me of countries with two mainstream political parties - neither discriminates clearly what their USP is...
They have a ton of fantastic consumer-level computing devices, and one ridiculously-priced mega-computer.
But there are many of us that want something in the upper-middle: a fast computer that is upgradable, but maybe $2k and not $6k (and up).
(The iMac Pro is a dud. People that want a powerful desktop generally don't want a non-upgradable all-in-one.)
Seems absurd.
I use MacStadium for compiling and testing iOS apps. I was wondering if the ARM machines would be worth a look, but they are disappointing. If I was still using Macs as my daily driver, I would buy the new MBA for a personal machine.
I was just window shopping a new gaming rig, and 32gb is affordable (100 bucks), 64gb (200 bucks). Cheap as shit, what’s the hold up?
Can you please provide the link to 64GB DDR5-5500 for $200? I'd love to buy some too!
I guess DDR 500 runs you 350. It looks like Apple charges you 600 for ddr 400 32gb. I don’t know, what am I missing here?
I bet they’re only upgrading it because it was essentially free. They already developed it for the developer transition kit.
I commend the idea of a small enthusiast mini desktop like a NUC but I don’t think the market is really there, or if they are, they’re not interested in a Mac.
or perhaps it's not 'paging', and just dumb luck I hit and see beachballs on multiple new higher-end macbook pros regularly.
And one tip is to use the right hand side USBC ports for charging, not the left hand ones as for some reason or other they tend to cause the machine to heat up more...
thx for tip.
On my machine it is true that charging on the right is better. Charging on the left spins up the fan.