• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Apple M1 CPU

People who buy Apple Laptop's are a different crowed to people who Buy Asus Laptop's, its not as if people who don't normally pay £1000 for a 13" Laptop are now going to on mass because Apple has a new CPU.

The only thing about it is Intel don't supply the CPU's anymore.

Well a decent amount of people might. The battery thing, is probably a bigger draw than the CPU processing IMO for that market. But then few people are travelling these days so perhaps it not the draw that it might have been pre-covid.
 
I mean they surely can, and already do to a degree, especially when they're backed into a corner. I just find Intel's marketing to be the most dishonest these days.

marketing has always been dishonest.

M1 is really as you say, first gen stuff in low-end products with limitations (very good but still integrated GPU, I/O and ram limits, etc). They said the transition takes 2 years and we're only 8 months into it. So I'm expecting a lot more stuff to come in the next 18 months.

I personally care about code compiling and M1 has been amazing at that, I'm only waiting for the 16-inch MBP with 64GB of ram and for docker to be properly released for M1 before asking my employer to buy one for me :D

The limitations, are what gives it part of its speed.

What do you compile out of interest?
 
There's nothing special about those limitations that increase speed.

There is. Its having everything close together on the one die, System-on-a-Chip or SoC, as little as possible data goes off chip. Less power needed, less heat etc. Smaller die.

Add more stuff will make it bigger.
 
There is. Its having everything close together on the one die, System-on-a-Chip or SoC, as little as possible data goes off chip. Less power needed, less heat etc. Smaller die.

Add more stuff will make it bigger.

Every modern CPU is a system-on-chip, has been for a very long time and it's been a multi-decade trend for the entire industry. In fact in terms of what's actually on chip, there's not much difference between M1 and any mobile Intel or AMD SoC.
 
Every modern CPU is a system-on-chip, has been for a very long time and it's been a multi-decade trend for the entire industry. In fact in terms of what's actually on chip, there's not much difference between M1 and any mobile Intel or AMD SoC.

There really is. RAM is on chip for example.
 
There really is. RAM is on chip for example.

Ram is soldered on package, not on chip. That's a huge difference. There's no commercial SoC with on-chip DRAM in the world, primarily because DRAM fabrication process is entirely different.

M1 uses exactly the same LPDDR4x DRAM that is used in dozens of other laptops, connected to the chip in exactly the same way. They're just soldered on the package instead of on the motherboard. They're still separate modules.
 
Which intel laptops have ram with this short a path the the CPU on the same substrate

Let me unpack this:

That's not what I said, you claimed M1 uses on-chip RAM, it doesn't, it only uses on-package RAM. Going from on-chip to off-chip (whether on-package or not) already adds almost all the latency anyway, the length afterwards (a few mm or a few cm) isn't that important and can be compensated with minor tweaks to the cache system.

You can look at Apple's own designs, A12X used on-package DRAM, A12 used off-package (PoP), and compare their SPEC benchmarks. Spoiler: there's no noticeable difference. Again, A14 used off-package (PoP), and M1 used on-package.

To see the effect of this, see variance of M1 vs Intel/AMD SPEC subbenchmarks (Anandtech has them) that rely on memory and those which don't. Spoiler: the variance is negligible. You can also see RAM bandwidth benchmarks of M1 compared to any AMD processors on similar memory, and it's pretty comparable.

In short, the "magic" is in the microarchitecture, not where DRAM is soldered. These stuff have been researched decades ago. If it was that simple, Intel and AMD would have done it decades ago. The reason they don't offer on-package DRAM is because they don't make DRAM and it just complicates their product for no noticeable benefit. Apple builds the whole thing, so if moving DRAM on the same package as SoC saves PCB space and simplifies the design, it's worth it for them.

What is actually different about Apple's RAM implementation is the unified architecture, which is different to AMD/Intel's shared architecture. In M1, different modules can access RAM that's assigned to another module (e.g. GPU doesn't need to copy information from CPU's portion of the RAM onto its portion of the RAM, it can just read/change it directly). This avoids a lot of data duplication in RAM. Meaning you waste less time copying data from one part of RAM to another, and you'll need less RAM to do the same task. However, that advantage will still exist if Apple decided to move RAM off-package and GPU off-chip (slightly more tricky, as Apple also shares cache with GPU).
 
Last edited:
Let me unpack this:

That's not what I said, you claimed M1 uses on-chip RAM, it doesn't, it only uses on-package RAM. Going from on-chip to off-chip (whether on-package or not) already adds almost all the latency anyway, the length afterwards (a few mm or a few cm) isn't that important and can be compensated with minor tweaks to the cache system.

You can look at Apple's own designs, A12X used on-package DRAM, A12 used off-package (PoP), and compare their SPEC benchmarks. Spoiler: there's no noticeable difference. Again, A14 used off-package (PoP), and M1 used on-package.

To see the effect of this, see variance of M1 vs Intel/AMD SPEC subbenchmarks (Anandtech has them) that rely on memory and those which don't. Spoiler: the variance is negligible. You can also see RAM bandwidth benchmarks of M1 compared to any AMD processors on similar memory, and it's pretty comparable.

In short, the "magic" is in the microarchitecture, not where DRAM is soldered. These stuff have been researched decades ago. If it was that simple, Intel and AMD would have done it decades ago. The reason they don't offer on-package DRAM is because they don't make DRAM and it just complicates their product for no noticeable benefit. Apple builds the whole thing, so if moving DRAM on the same chip as SoC saves PCB space and simplifies the design, it's worth it for them.

What is actually different about Apple's RAM implementation is the unified architecture, which is different to AMD/Intel's shared architecture. In M1, different modules can access RAM that's assigned to another module (e.g. GPU doesn't need to copy information from CPU's portion of the RAM onto its portion of the RAM, it can just read/change it directly). This avoids a lot of data duplication in RAM. Meaning you waste less time copying data from one part of RAM to another, and you'll need less RAM to do the same task. However, that advantage will still exist if Apple decided to move RAM off-package and GPU off-chip (slightly more tricky, as Apple also shares cache with GPU).
Well said.
 
I personally care about code compiling and M1 has been amazing at that, I'm only waiting for the 16-inch MBP with 64GB of ram and for docker to be properly released for M1 before asking my employer to buy one for me :D

Out of curiosity have you got any links to decent comparisons/benchmarks looking specifically at code compilation?

I would've thought that, looking at current chips at least, the 8 cores with no SMT vs 8 cores/16 threads would hurt it? In my experience the SMT boost for compilation is pretty good. But happy to be proved wrong :)

Of course if they add cores when they do the more powerful stuff then that's a moot point...
 
Every modern CPU is a system-on-chip, has been for a very long time and it's been a multi-decade trend for the entire industry. In fact in terms of what's actually on chip, there's not much difference between M1 and any mobile Intel or AMD SoC.

Ok I stand corrected. But what Intel laptops (ignoring ARM) have similar short path to memory. Even if it makes negligible difference to performance and its all around the unified memory.

I'm also curious if the gains will be maintained if they build a Mac Pro level machine for professionals or even gamers. I'm also thinking of how they are looking at ARM for servers. Will this Apple architecture be applicable there.
 
Out of curiosity have you got any links to decent comparisons/benchmarks looking specifically at code compilation?

I would've thought that, looking at current chips at least, the 8 cores with no SMT vs 8 cores/16 threads would hurt it? In my experience the SMT boost for compilation is pretty good. But happy to be proved wrong :)

Of course if they add cores when they do the more powerful stuff then that's a moot point...

Compiling is single-threaded, parallelism is achieved by just compiling different files at the same time (which is very common and scales well, assuming the project isn't huge one huge file). Lack of SMT is also not a disadvantage in general, really depends on microarchitecture. Apple is much more aggressive at out-of-order execution and is better at branch prediction, so they can utilise a higher percentage of a core running a single thread, improving single-threaded performance and reducing the need for SMT.

TechCrunch compiled WebKit (which is C++) as part of their benchmarks and the 4+4 core M1 beat 8-core/16-thread MacBook Pro and was as good as the 12-core/24-thread Mac Pro.
https://techcrunch.com/2020/11/17/yeah-apples-m1-macbook-pro-is-powerful-but-its-the-battery-life-that-will-blow-you-away/

These are also Java, python, Go, SQLite, Redis and JavaScript comparisons, although with Intel 9th gen and Ryan's 3800X, again M1 is just amazing at doing these stuff:
https://tech.ssut.me/apple-m1-chip-benchmarks-focused-on-the-real-world-programming/

Also lots of articles like this people just comparing their own code compiling workflow:
https://build2.org/blog/apple-m1-compilation.xhtml
 
Ok I stand corrected. But what Intel laptops (ignoring ARM) have similar short path to memory. Even if it makes negligible difference to performance and its all around the unified memory.

I'm also curious if the gains will be maintained if they build a Mac Pro level machine for professionals or even gamers. I'm also thinking of how they are looking at ARM for servers. Will this Apple architecture be applicable there.

The microarchitecture is suitable for severs (the tasks are not that different, Intel, AMD and ARM also share their microarchitecture with across multiple product lines, and SPEC benchmarks are industry standard for server chips). Apple changed the macOS license to allow them being leased out by cloud providers and AWS is offering them but currently mostly for software development purposes. Apple seems like they have an eye for this market. Although I don't expect much to happen for at least a few years.
 
Compiling is single-threaded, parallelism is achieved by just compiling different files at the same time (which is very common and scales well, assuming the project isn't huge one huge file). Lack of SMT is also not a disadvantage in general, really depends on microarchitecture. Apple is much more aggressive at out-of-order execution and is better at branch prediction, so they can utilise a higher percentage of a core running a single thread, improving single-threaded performance and reducing the need for SMT.

TechCrunch compiled WebKit (which is C++) as part of their benchmarks and the 4+4 core M1 beat 8-core/16-thread MacBook Pro and was as good as the 12-core/24-thread Mac Pro.
https://techcrunch.com/2020/11/17/y...its-the-battery-life-that-will-blow-you-away/

These are also Java, python, Go, SQLite, Redis and JavaScript comparisons, although with Intel 9th gen and Ryan's 3800X, again M1 is just amazing at doing these stuff:
https://tech.ssut.me/apple-m1-chip-benchmarks-focused-on-the-real-world-programming/

Also lots of articles like this people just comparing their own code compiling workflow:
https://build2.org/blog/apple-m1-compilation.xhtml

That's some impressive performance :)
 
Those are some pretty solid performance figures.

Yeah I think I'm just about sold on one of these, probably the Air, especially with the DisplayLink 'trick' to get more screens working (bar a couple of niggles like HDCP). My old (late 2013) 13" Macbook Pro Retina is getting a bit long in the tooth now.
 
Those are some pretty solid performance figures.

Yeah I think I'm just about sold on one of these, probably the Air, especially with the DisplayLink 'trick' to get more screens working (bar a couple of niggles like HDCP). My old (late 2013) 13" Macbook Pro Retina is getting a bit long in the tooth now.

I've had my M1 air since late November and absolutely love it. Battery life is great and I've not had it stutter once. That and being completely silent is a bonus. The only issue I've had is intermittent bluetooth dropouts
 
I've also had my M1 air since launch, and I love it. Instant on makes me reach for it in cases where I might have reached for my phone previously and the battery lasts so long, I've never found myself out of battery.

It's the perfect daily/general laptop.

What I'm waiting for is better support for ARM/QEMU from the various docker image vendors I use during development, to see how my development workflow is on the M1 vs my usual Linux desktop. I'm hoping QEMU is an improvement as Docker on Intel macs was painful.
 
Keep an eye on your SSD read write numbers. I'm seeing some people mentioning that M1 Macs have extreme swap usage, like nearly a TB of writes in 2.5 days
 
Back
Top Bottom