> a poor understanding of computers would be to think that speed is all that matters.
> What good is 16GB of super fast RAM if the machine is swapping out to disk regularly?
If the swapping is fast enough not to be noticeable, then speed is all that matters. In fact, if SSDs were about as fast as RAM, we wouldn't need RAM at all!
The whole reason why we have all these layers of memory (disk, RAM, L3, L2 and L1 cache, etc) is exactly speed, nothing else. The closer to the CPU, the faster, but also more expensive.
Apple must've spent a fortune on marketing bamboozlement ("you don't need as much RAM with apple silicone") that even people on HN now seem to parrot it.
NVME is around two orders of magnitude slower than DDR3 RAM, so as soon as your heap is tapped out you'll hit a performance wall.
As I sit here and type this, my 10.15.7 Catalina desktop is sitting at 12.25Gb of used memory with 2 Edge tabs and an open Citrix session. Any actual, professional use will put you way north of even 16gb.
> as soon as your heap is tapped out you'll hit a performance wall.
Yes, probably. Two points though:
1. My comment was that _if_ SSD performance was comparable to RAM, there would be no need for the latter.
2. I have very rarely swapped on my 16Gb M1 Air, and when I did I only noticed later when looking back at graphs. I'm sure there was a performance hit but I never felt it.
> Any actual, professional use will put you way north of even 16gb.
I do plenty of actual, professional use on my 16Gb M1 Air. Right now I have 3 VS Code windows open compiling Go code, running acceptance tests and whatnot. I also have Mail, Safari (with tens of tabs) and Firefox (with >100 tabs) open. A few minutes ago I also had Slack open on it. I regularly start iTerm to do terminal tasks.
You must have an Intel machine. I had a powerful fully specced 32Gb i9 16" MBP. This cheap and humble 16Gb passively cooled M1 blows it right out of the water in every single aspect. It's even better at running Intel Docker images!
I run an 8GB M1 and rarely hit the wall, and I use mine much like you describe using yours. I haven’t yet cared that I have 8GB, and I was expecting to have a rough time with it (this is sort of a stand-in until I decide what I want longer term - these are incredible machines).
>Any actual, professional use will put you way north of even 16gb.
Well, I use my M1 (16 GB) for heavy video editing and music sessions, plus professional programming, with IDEA, VMs, and so on. And I don't seem to ever need even close to 16GB, much less "way north", or even saw it be slow.
>As I sit here and type this, my 10.15.7 Catalina desktop is sitting at 12.25Gb of used memory with 2 Edge tabs and an open Citrix session. Any actual, professional use will put you way north of even 16gb.
This is the very type of bamboozlement I was alluding to - instruction set and/or silicon architecture doesn't actually change the amount of data you use.
Citrix will still need to buffer the same (compressed?) 4K worth of pixels and Edge will still to load up the full DOM, cache all the sources, stand up a sandbox with javascript virtual machine etc.
There's nothing magical there... a single frame of 16 bit 8K RAW will allocate the exactly same amount of heap on M1, M2, Intel or anything else for that matter.
I'm now playing a 4K movie on Plex. It's using a little over 1Gb RAM for that. After I started it, my computer swapped ~250Mb to the SSD. Firefox is still snappy. Safari is still running normally. VS Code is still doing its thing. I even started a `brew update` on iTerm, and I have the App Store installing a couple apps. Mail still runs fine, sends and receives, and I can switch between mail accounts and messages.
So, you may be technically right about the memory usage and the need for swapping. What you're missing is the fact that it doesn't hurt the user experience.
This is not marketing or bamboozlement. This is me on the same laptop I'm typing this answer.
You should probably do a deep inspection of your beliefs and stop denying the actual, practical experience of many people who are responding to you with real world experience.
> This is the very type of bamboozlement I was alluding to - instruction set and/or silicon architecture doesn't actually change the amount of data you use.
Oh, but it does when it has a hardware memory compression engine. The very different GPU design also means it can use less (or more) memory in different situations.
Yes, "a single frame of 16 bit 8K RAW" will be the same size expanded. But on M1 it can be shared more easily with different parts of the CPU, GPU and co-processors without copying.
Even more so for other pipelines, involving CPU+GPU.
MacOS also has memory compression, assisted by the CPU.
And there are other ways an OS can use to keep memory usage lower given a different CPU architecture...
> NVME is around two orders of magnitude slower than DDR3 RAM,
No. Dual channel DDR3 went up to about 18 GB/s read and 14 GB/s write. [1]
The latest NVMe PCIe 5.0 SSD are about 13 GB/s read and 12 GB/s write. [2]
Apple SSD are about half that now, but most DDR3 users didn't have highest clocked dual channel RAM either. So roughly a factor of 2 or so at best, about a 50x difference compared to two orders of magnitude :-)
Ooooh good point - haven't actually heard of newest NVME stuff before.
The M1/M2 apple hardware specifically seems to be around an order of magnitude from what I've seen (ddr4x 60-70gb/sec vs 7gb/sec for the NVME or thereabouts).
The obvious observation here is that the fab yields for high memory apple silicone must not be all that great, which is why they're mostly shipping 8 and 16gb versions.
Never mind comparing bleeding edge NVMe with (by now) decade-old DDR3, while the current bleeding edge DDR5 is now trickling out at 100GB/s+ pretty easily.
Never mind DDR memory latencies of ~50ns, vs NVMe at 50uS+.
Just because the user doesn't notice GUI problems doesn't mean it's not going to be a catastrophic bottleneck in any memory-intensive application.
I'm sitting here on an M1 MacBook Pro with 64GB of memory. I'm running Chrome with a few dozen tabs open, Slack, VS Code, and Terminal. Apparently 32GB of memory is being "used".
Do you believe I would experience noticeable performance loss if I was running 16GB of memory?
Well yeah - getting that extra 16gb to the CPU from your SSD would take you around 4.5 seconds, versus 250 milli from RAM. Not to mention random access latency, nanos vs mics.
Over 22GB of that is cache. I’m sure some of that is useful but the overwhelming majority of it is just being used because the memory is freely available so why not? If it improves performance it’s not in anything I’ve ever managed to notice between my 16GB work machine and 64GB personal one (which are used for largely similar tasks, though I’d say the work one is a bit more heavily stressed).
Sure some cleanroom laboratory benchmark will say the 64GB version is faster. And for some workloads I’m sure it’s a bigger deal. But a human just isn’t going to perceive much practical benefit from a full Chrome restart being 5% faster or whatever.
> Any actual, professional use will put you way north of even 16gb.
Plenty of people will be more than happy with 8GB of RAM on an M1. I had a base model M1 Mac Mini and it worked great even for gaming.
Obviously if you're running VMs or editing video or doing any other memory intense workload you're going to need more memory. But for anyone who doesn't need an absurd (>16gb) amount of memory (college students, many programmers, those who use computers just for web browsing and Netflix), 8GB or 16GB will feel snappy.
In fact, if SSDs were about as fast as RAM, we wouldn't need RAM at all!
Just in case anyone was curious, I looked into this recently, assuming 7-8GB/s was pretty darn fast, but dual channel DDR2 RAM was doing such speeds in 2002. So a high end SSD is about as fast as 20 year old memory, if you ignore latency.
I don't know what people are reading into my comment. I said _if_ SSD speeds were comparable to RAM, RAM wouldn't be needed. The reason we need it is precisely because disk is slower and cheaper.
Destroying the resale value of your laptop because trashing through a non-replacable SSD is not something I would really appreciate. We don't have(?) official numbers, but consumer SSDs go as low as 150TBW
You’re describing your results on an Intel Mac. They aren’t relevant to this particular bit of the discussion. And, owning two different M1 laptops, I concur that the relationship between the RAM and SSD and M1 SoC yields fundamentally different results from prior hardware. The result is stunning.
What relationship would that be?
The DDR4x ram in the M1 can hit 70gb/sec, compared to 7gb/sec and ~3.5gb/sec average for the NVME. There's no amount of magic to get around those facts.
I'm a professional systems programmer. I agree with everything you're saying, in principle.
Except I have two M1 MacBook Pros (one 16GB, one 64GB). I frankly don't notice the difference between the two on any workload. This involves compiling multiple projects, running multiple docker containers locally in virtual machines, running VS Code, Slack, Chrome, and other productivity tools.
These machines frankly feel like alien fucking technology and I don't say that lightly. I'm used to year-over-year improvements being almost unnoticeable. These machines feel like I've jumped ten or more years forward in performance and responsiveness. And they do so while barely generating heat. My previous Intel MBP would make my home office hot just by being in a Zoom call. When I first got my M1 MBP, I left it running all night performing a SIMD-heavy pure math workload pegging every core at 100%. Not only was it faster, but the room wasn't even perceptibly warmer than ambient.
I have a gaming PC next to my work area. Zen 9 3950X, GeForce 3080, 64GB of DDR-4000 RAM, and Samsung 980 Pro SSDs. The fans go blazing and I start sweating just booting Windows. And unless I'm doing something in VR, it performs comparably to the MacBook Pro. I barely turn the damn thing on these days.
The RTX 3080 is a massive power hog. All that energy has to be dissipated somehow.
It's really impressive what Apple has managed to do with the M1 and M2 chips. If an Apple M chip ever offers comparable performance on AAA games, I'll get rid of my desktop computer.
What does the OS do if the application demands more than 8gb of memory? I have use cases in the 14gb memory range for a single threaded application. Would the application crash since its not built to anticipate swapping?
If I go to their store website it shows me only two options:
- 8 core, 8 core gpu, 8gb, 256
- 8 core, 10 core gpu, 8gb, 512
No indication whatsoever that this 8gb is not fixed. Nor is the storage fixed. But the only way to find out is to try and buy it, at which point I can just turn the first model into the second or something else entirely. Also, none of the presented options is anywhere close in price to the top spec model.
True, in the German store it says "Wähle deine Konfiguration beim nächsten Schritt."
> how would you know that ram is confugurable?
2 options:
- Scroll down a little, and under "Gemeinsamer Arbeitsspeicher" it says:
"Der M2 hat einen schnelleren Arbeitsspeicher als der M1, mit 50 Prozent mehr Speicherbandbreite. Er kann mit bis zu 24 GB konfiguriert werden – 8 GB mehr als beim M1 – für flüssigeres Multitasking und einfaches Arbeiten mit großen Dateien."
- Click on the "Wählen" button under the chosen model. It's very obvious in the next screen that you can configure RAM.
> It's very obvious in the NEXT screen that you can configure RAM.
That is the whole point and I said that already...
If you look through this thread you see quite a few peopke thinking you can only get 8gb of ram because the only two models shown feature 8gb and no text next to that indicating otherwise. Only if you think that apple might be lying, you go to the next page and see that they offer more than 8.
> Scroll down a little, and...
No, scroll down a lot and then it is hidden in the fluff text. Great, doesn't change that it is missing in the central place people actually look at.
All the "well technically you can find out by jumping through these hoops" does not change that this is bad UX.
> What good is 16GB of super fast RAM if the machine is swapping out to disk regularly?
If the swapping is fast enough not to be noticeable, then speed is all that matters. In fact, if SSDs were about as fast as RAM, we wouldn't need RAM at all!
The whole reason why we have all these layers of memory (disk, RAM, L3, L2 and L1 cache, etc) is exactly speed, nothing else. The closer to the CPU, the faster, but also more expensive.