After upgrading our app's dependency chain to run native arm64 builds (which took a bit of googling) webpack's incremental build time on our app is down to 116ms from 983ms on my 2016 Macbook Pro. Over 8x faster.

Our Tensorflow.js (webgl) models that previously ran about 8fps in the browser are now running at the camera's native 30fps framerate. (Similar to my desktop 2080ti.)

(FWIW, my WindowServer usage is generally around 3-8% when idling and I am losing about 6% an hour at the moment.)

I think I want to just use this machine fresh and just move a few files over as needed.

There was a bug where something would infinite loop on a non-standard resolution, think that was fixed.

So glad I went with the spec I did, all the power I need with room to grow.

I don't think people will be moving away from webpack for awhile, but transpilation is often a very expensive part, and `swc` is very near to stable, and as a transpiler / polyfilling tool it's pretty fast.

Because buying new stuff is wasteful and only a temporary solution to the problem?

Look, I like nice, fast machine as much as the rest of the crowd here, but let's not kid ourselves. In a years time the software we run on it will have become even more bloated and things will be just as slow as they were. And then a new machine comes out we all want to buy that.

There will be some ecstatic blog posts again. Which is weird, I think. Needing a faster machine for you daily work should feel like defeat for a developer.

If you need a new laptop and can afford it, by all means go ahead and indulge. In the mean time, I am typing this from a 7 year old Latitude that I use every day for software development and that shows no sign of aging, apart from some scratches. And each time I read a post like this I wonder: what do these people _do_ with these machines?

To give you an idea, I don't have Macs, but I have a beefy desktop because I run my own OpenStreetMap server on it.

My old machine: i7 3770, 32GB ram, 1 TB sata SSD: 3 days to import Europe's map. The new one: Ryzen 5800, 32GB ram, 1 TB NVME: 10 hours to import Europe's map.

I will probably need to upgrade my ram and NVME capacity if I want to build a world map.

What do you do with it?

Enjoy your latitude, but you don’t have the right to tell me that I can’t enjoy my M1 max.

Different strokes for different folks.

Would you mind sharing your research results?

I am getting one soon and I am in progress of drafting notes for stuff like migration & setup.

We were still on node 12 because, until recently, Google Cloud Functions didn’t support anything newer. But when I installed it, I noticed it was running on x86 via Rosetta. So I upgraded to the latest node (17) which installed as native arm64.

Almost everything “just worked” but there were build errors with installing: webpack, node-sass, gprc, and node-canvas.

Webpack turned out to be an issue with node current (17) so downgrading to lts (which is currently 16) worked.

That seems to have magically fixed the gprc build too.

We weren't actually using node-sass anymore; so removed that dependency.

Upgrading node-canvas to 2.8.0 and building from source (after installing its dependencies via homebrew) seems to have worked.

All in all it took maybe 2 hours to follow all the rabbit holes. And our whole engineering team is going to move over to arm64 as soon as Apple can deliver their machines.

I’m hoping for a mid-tier unit though - a rebirth of the G4 Cube as a “Mac Mini Pro”

I daily drive an M1 mini now and abuse the hell out of it. I’ve never heard the fan. It’s a remarkable device. I was installing a Homebrew package the other day on my i7 MacBook Pro and the fans kicked on - it almost startled me - I had forgotten about active cooling.

Those sockets need to communicate with each other, and extremely quickly. And that I/O is very, very expensive in terms of power. Even with that you still end up with poor performance scaling, especially with anything that isn't numa aware (which almost nothing is)

So no, that extrapolation doesn't make sense. But even ignoring that missing cost, 8x M1 Max's would be ~800w, give or take. That would work in a Mac Pro replacement, but that'd still be a significant power increase over many of the existing models.

Here's EPYC 7601's "IF" power (it's really all the uncore power) vs. the core power.

Going outside the chip means swinging lots of capacitance up and down, and modern interconnects swing that capacitance fast.

That will be one of those shut up and take my money moment. Given how they have been going retro or paying tribute to past design in their recent new product lineup I am hoping a G4 Cube revival is actually within a possibility.

Replicating the capabilities of the Mac Pro with would require another model of M1 and for the volumes the Mac Pro sells I do not believe developing a dedicated lithography mask makes sense. Besides, most of the unique capabilities of the Mac Pro are divergent with vision of the M1:

- PCIe slots, there are no GPUs compatible with Arm Mac and some of the functionality of the Afterburner card has been integrated into the M1 Max

- Lots of RAM, as seen with the lower RAM models of Mac Book (Air) Apple is increasingly relying on swapping to the fast SSD.

But performance, even with half the ram, was still stellar. Everything remains speedy, even when it's doing far too much.

And I don't think Apple would even want to be competitive in this segment.

Would love to have reasonably costed m1 instances but the value play for us is ampere.

So, what you're left with, is a 64 core M1. Which is still going to be less efficient than a Threadripper, an EPYC or even a Xeon (and that's quite sad) at multithreaded tasks, and, knowing Apple, about twice as expensive.

For example, a piece of code I was optimizing recently ran with very little variation in time between runs on my M1 system, but had ~30% variation on a similarly loaded Intel Mac. Changing the QoS level to Default instead of Utility allowed the Intel Mac to perform much more consistently on the benchmark, but made no difference on the M1 (I have good reason to leave it at Utility in my actual application, but for benchmarking it made sense to try other levels).

I found it hard to tell from the article, but it sounds like on the M1 Pro maybe the two highest QoS levels (interactive and user initiated) now preferably map to p cores 0-3, and maybe now default and utility map to p cores 4-7, and then background gets the e cores?

> it sounds like on the M1 Pro maybe the two highest QoS levels (interactive and user initiated) now preferably map to p cores 0-3, and maybe now default and utility map to p cores 4-7

That seems unlikely, though I don't have an M1* device what feels more likely is that the two clusters are independently powered (the turbo-ing works on a cluster basis so that's almost certain), therefore the machine favors fully loading the first one before it starts "spilling" to the second (and thus has to power it up).

With that said, there is a huge M1 hype on HN. It reminds me of the 64-bit ARM hype when someone benchmaked encryption (because ARMv8 has AES instructions) and everyone were made to belive that the new iPhone would be 3-4 times faster.

The main thing driving this is that people are comparing a new M1 to their Intel Mac laptops with processors from ~2016 and then naturally it's a marked improvement because it's being compared to a five year old machine.

You compare them to a modern Intel or AMD system and they're competitive. They win some and lose others. Much was made about the M1 Max taking the single thread performance crown; for less than a month. Golden Cove from Intel just took it back. It's plausible that it will be AMD's again in a couple months when we get Zen 3 with 3D V-Cache.

They're all within single digit percentages of each other and the winner is whoever released their top end processor most recently.

This is quite misleading in multiple ways.

The M1 Max has the same single thread performance as the M1, which has been out for around a year.

You're also comparing Intel's latest desktop CPU to a laptop CPU. Before the M1, a laptop CPU competing with the highest-end consumer desktop CPU in multi-core OR single-core was unthinkable and had never happened. Yes, there have been leaks in which Alder Lake's mobile CPU beats the M1, but we don't know whether these benchmarks are representing the numbers we'll be seeing in consumer laptops with laptop cooling.

That being said, I hope there will be competition!

[0] almost literally by the thermal numbers

I compared my 2014 MBP to 2019 MBP and there was no marked difference. It was running hot and eating battery very fast. 2019 Intel Mac and 2020 M1 Mac are two completely different animals. Also, the price drop for a better-performing machine was jaw-dropping.

The i9-9980HK (released April 2019) in the 16" 2019 Macbook Pro looks like the highest-end mobile i9 Intel had released as of November 2019 when the laptop launched.

And the i7-9750H in the base model was also launched in April of that year, and looks like it was their 2nd-fastest mobile hexacore i7 at the time?

I don't think the GPU is much of an upgrade, but it does have tensor cores, the better video encoder (could be an OBS monster) and I believe better thermals. I am sure putting Pop!_OS or Manjaro on it would make this a great *nix platform for developers, content creators or streamers.

Trying to find a cheaper apple quality device that is as good... you'll really be limited to the cheaper Yoga 4k screen models and their ilk. But I have no experience with those devices.

This is the inconvenient truth of these discussions. If anything, Apple is under playing the bang for the buck and performance per watt of these machines.

All the Intel machines I've looked into have fans and still manage to be 30-50% slower.

I’m fact I use an M1 iPad right now. It’s great. For browsing Facebook and video calls and doing remote work, it’s perfect. And for serious workloads, those go straight to my desktop workstation (AMD). Best of both worlds.

I mean, yes, somewhat, in (almost comically) high power desktop configurations, depending on which benchmark you believe. This is of cold comfort to anyone who wants a fast laptop, though.

To get the most energy efficiency out of the M1s is to use binaries/programs that have been specifically compiled for the M1.

M1 macs are just really damn fast for daily usage.

My M1 Macbook Air blows my 16" out of the water in every conceivable way. I'd elaborate but there seriously isn't any dimension of it that isn't an improvement. Maybe the smaller display?

That is the issue though. Intel macbooks were notoriously badly cooled. That hurts the performance a lot. If all you have used is that, of course you find the new one a lot better. However, compared to any other high end Intel/AMD laptop, you would not perceive this much difference in speed.

Pick up a Macbook 16 and XPS 15 from 2019 with the same specs (CPU/RAM) and the XPS absolutely runs circles around the macbook when it comes to general responsiveness in UI. All because of cooling.

One visit to XPS owner’s subreddit was enough for me to stop entertaining the idea of buying it.

if i run same vscode dotnetcore workload on m1 air, that thing lasts atleast 10hours on battery and no heat whatsover. better, the compile time is usually faster than 3800 bucks worth dell precision (on a 1200 bucks laptop)

First off the M1 Air has no fans.

Second the MBP 16" was the one with the improved cooling, so it wasn't throttled randomly by using the wrong charging port.

And finally the whole reason the M1s are so well cooled is there's no 100W SoC, or 200W probably when you consider the additional dedicated GPU in my 16".

My M1 has a 10W TDP!

10!

I don't think the GPU alone in my 16" could have been cooled by a solution handling 10W.

-

Also note: in your other comments you keep mixing TDP and wall power draw, my 16" would have had higher wall power than my M1 Air by far.

Even the M1 Mac Mini maxed out at like 30W from the wall: https://images.anandtech.com/graphs/graph16252/119344.png

And my M1 has a lower rated version of that (10W vs 24W)

For desktops I don't care about TDP but a laptop is the exact opposite.

Your M1 Air doesn't have an M1 Pro or Max, either, which I clearly specified. Why are you talking about M1 power draw in your response when that's not the SoC in question?

> Also note: in your other comments you keep mixing TDP and wall power draw,

Power in == heat out. CPUs don't do mechanical work, all power is converted to heat.

If the 'TDP' is less than peak power draw (and is actually respected in some form), it simply means the SoC is intended to power throttle after a duration. But that's so far not the case for any of the M1's, so claimed TDP is fully irrelevant as it's seemingly not enforced in any way. Which isn't unique to Apple fwiw, AMD's claimed TDPs are equally irrelevant.

You're free to bring in your third option which clobbers it even further, but an $800 fanless machine also clobbers it so it's a perfectly valid comparison when trying to counter weird claims that Apple only just now started working on proper cooling...

tl;dr: How can you claim the old 16" is only being beat because Apple just now started to care about cooling when a fanless machine also beats it? -

Also TDP is an advertising figure but when you're talking about orders of magnitude it's a perfectly fine way to talk about parts as long as you stay consistent in TDP vs wall draw

Wall draw to wall draw, my M1 Air draws under 30W and beats a 16" drawing over 100W

TDP to TDP my M1 is 10W to a 16" closer to 100W.

In both cases it should be obvious why cooling is less of an issue...

This is completely incorrect. You cannot compare TDPs in any meaningful way except kinda within a single vendor's single generation. There's no standard definition for TDP, so it cannot be generally compared across brands.

And for most brands TDP doesn't have any useful meaning, further eliminating any value to be had from it.

> Wall draw to wall draw, my M1 Air draws under 30W and beats a 16" drawing over 100W

Are you sure? What workload(s)? How did you measure power draw? And 16" isn't a CPU/SoC, so what are you even comparing against? Your M1 Air sure as shit isn't beating the current 100w 16" after all. So I'm guessing you're comparing against a previous MBP 16" of some unknown variety, which one specifically in which configuration?

> TDP to TDP my M1 is 10W to a 16" closer to 100W.

Apple never shipped an Intel CPU with a 100w TDP in any MBP of any size. So not only is this comparison meaningless, you're not even using the right numbers. The i9-9980HK that I'm guessing you're referring to is a 45w TDP. But see above about you can't compare TDPs, they have no meaning.

The older ones got really toasty, but it's not like the XPS faired better in the thermal design department.

Both these had their own quirks, (I returned the XPS I got from my place of work twice because I couldn't stand the insane amount of coil whine)

Been running the 2019 MBP since it was released and it's been a pretty pleasant experience overal. Only issue I've had was with external displays drawing too much power, which was fixed by forcing the refresh rate to 59.98Hz.

The previous setup was closer to 200W SoC (100W each for CPU and GPU) than 100W total.

From Anandtech[0]:

> Finally, stressing out both CPU and GPU at the same time, the SoC goes up to 92W package power and 120W wall active power.

2. Yes, you can burn the battery quickly under heavy use. I've killed it easily in about 2 hours during heavy Unity work.

3. When I was doing that, I did experience heavy fan output, about as loud and as hot as my 2018 15" MacBook Pro. After the heavy utilization stopped, it cooled down much more quickly than the older Intel machine.

[0] https://www.anandtech.com/show/17024/apple-m1-max-performanc...

[1] https://www.notebookcheck.net/Apple-MacBook-Pro-14-2021-M1-P... , in the "Battery Runtime" and "Power Consumption" sections near the bottom.

In a 16” at least, the Max can go up to 92W per anandtech’s initial test, and display + ram + rest brings the wallplug draw to 120.

I have a hard time believing that one company is doing miracle work in a small and light laptop. That's also true in the other direction - I don't really believe that Apple's machines deliver desktop performance on battery remaining cool and quiet over longer times...

I hope they improved in 2019, mine was only from a year before that.

I wonder how much of that is due to a "finally, a good laptop" sentiment. I have used a lot of laptops over the years, and each and every one of them have been terrible in lots of ways, especially recently. If M1s "solve" the laptop problem, that would be great.

It's less powerful than a modern desktop, but not by much. More importantly, none of the power management, suspend, wifi, whatever... ever seems to fail.

UI scaling is also a big one. I have a Dell XPS 15 in 2014/2015 with Windows 8.1 and the UI scaling wasn't great.

other tools like the Dash app are also amazing tools to speed up the development process. It will take some getting used to moving from Windows and to a far lesser extent linux, however with the right tools for the job you will end up being more productive and less frustrated. Especially compared to Windows.

Unless I need something that specifically only runs on Windows, I would use Mac/Linux over it any day of the week. For productivity it's Mac > Linux >>> Windows for me.

and https://support.apple.com/en-us/HT204948

A lot of the pthreads books I see are from the late 90s. Is there a more recent reference? What's the best way to write cross-platform (e.g. not Grand Central Dispatch) multithreaded apps with these new chip architectures?

They move everything that isn't foreground to an efficiency core, which is awful for compiling or video processing.

There's apparently a BIOS option that will use ScrollLock for disabling the efficiency cores entirely.

This is front of mind for me since reading a Cloudflare blog regarding AVX-512 instructions invoking dynamic frequency scaling to manage power/thermal capacity on chip. (https://blog.cloudflare.com/on-the-dangers-of-intels-frequen...)

If this is happening on Xeons, it's probably happening on consumer dies as well, in addition to other non-obvious power/performance optimizations. Perhaps this is why Alder Lake is pumping up the TDP[1]?

edit: [1] https://news.ycombinator.com/item?id=29106860

Windows has had that (foreground boost) for a long time, Intel probably piggybacks on it. It'll be interesting to see how it will behave on Linux, which AFAIK never had that mechanism (except perhaps on Android).

For the Mac, I believe you have equivalent access for scheduling between posix and GCD, but the scheduling configuration is likely way more approachable in GCD.

Also: On M1, there is an added capability to run in a stricter memory model to speed up x86_64 emulation. This only is available on the performance cores, which is one of the reasons people observe non-native code draining the battery quicker.

But as to the latter assertion, you're indeed correct per Joe Groff (Swift compiler engineer at Apple): https://twitter.com/jckarter/status/1332045390057639939

> The A12 only supported TSO on the performance cores. The M1 supports it on all cores.

https://scholar.google.com/scholar?hl=en&as_sdt=0%2C33&q=asy...

Eh, maybe. I'm not inclined to rely on the reported % CPU usage to represent anything useful. For example, it's very easy to be at 100% CPU when you're actually spending most of that time waiting for memory reads.

I have seen on linux that I can be running something that uses 100% and the cpu is at 50c, then I run prime95 also at 100% and the cpu instantly hits 99c

It's a simple metric, but it can be complicated to relate it to what is going on with execution because the real system is very complicated.

The CPU % metric does not take into account execution strength, of the CPU, big vs small, dynamic CPU frequency, or effect that one CPU might have on another (e.g., SMT or shared caches or memory controllers).

Further complicating it is that the Linux CPU scheduler is not work-conserving. So you could have at least 100 application threads runnable at any given moment in your workload, but your 64 CPU system might only be hitting 80% CPU busy.

Then you get to application and kernel effects of course, locking, blocking, etc. can mean you don't even have as many runnable threads as you might think.

Then how all this actually relates to the work and heat the system creates is another matter again. Simple integer execution might only use half the CPU power of a vector heavy workload because you have fewer transistors clocked or switching (or even powered) in the core.

Consider this: 1. 14” or 16” custom miniLED display 2. Anodized one-piece aluminum clamshell 3. Completely custom SoC (much more expensive than buying Intel’s latest, at least in the short-term) 4. Completely custom OS with a full suite of native apps 5. Custom MagSafe adapter 6. Thunderbolt 4 (3 of them)

Now consider that this laptop outperforms most desktop workstations that cost $4k+…and does it on 16h of battery life.

If you’re in the market for a high-performance computer, you’d have to be dense to not buy one of these (or building x86 firmware)

> If you’re in the market for a high-performance computer, you’d have to be dense to not buy one of these (or building x86 firmware)

I'd love to see which $4000+ desktop gets beaten by any M1 processor. $4000 is A LOT for a desktop computer. I paid (in total, with upgrades over the years) around ~$1600 and easily beats anything provided by Apple.

I don't think so, quick comparison, with numbers from https://www.cpubenchmark.net/compare/Apple-M1-Pro-10-Core-32...:

- Apple M1 Pro 10 Core 3200 MHz - Score: 23730

- AMD Ryzen Threadripper 1950X - Score: 27293 - Price: ~$499

And I just spent 2 minutes digging up those numbers, didn't even look very carefully.

All the recent M1 benchmarks from respectable news sites confirm the same for the majority of multi core real workloads.

tl;dr you absolutely will not beat M1 Pro with a $1500 PC for work.

> https://www.cpubenchmark.net/singleThread.html

> $1000+ territory for the CPU alone to get close to M1 multi-core performance

Your claim was that you need a +$1000 CPU to get close to M1 multi-core performance, then now you link the single-thread tests?

Here’s a full comparison between M1 Max and the 1950x: https://browser.geekbench.com/v5/cpu/compare/10837819?baseli...

The M1 Max is ahead by leaps and bounds on many tests despite the synthetic score being close. And these only tell half the story. In all the reviews I’ve seen so far, it consistently beats the iMac Pro by up to 2x-4x in real word tasks. Mind you, the original M1 chip already beat many high-end desktops, so this is not surprising at all.

Even if the cpu was that cheap: $600 cpu + $200 mobo + $100 ram + $100 ssd + $50 power supply + $600 rtx 2060

That’s the best machine you can put together with ~$1500 and it will be absolutely smoked by the M1 Pro/Max. Then you can try to find a mini-led 4k 120hz monitor, peripherals, mic and speakers, all for < $1000 (hint: impossible)… and still not be able to run on batteries or pack it in a bag.

Don't be disingenuous just to win the argument.

To actually prove your point, you need to look at GPUs. A 3070 is ~$1100, so there we go, already we've broken his $1600 budget, and we have bought none of the other parts (which would total at minimum ~$1000 more).

If you go 5900x plus 3070, you're pretty much right in line with M1 Pro, but it will cost you around $2500 to equal the Macbook pro. So an upcharge of $1000 to $1500 for the price of being a mobile workstation, as well as all the Mac software. Not really unreasonable.

I'd really want to see that setup and benchmark

That might've been for when it was only an 8GB increase in RAM.

This sounds very unlikely? Apple has been building their own silicon for a decade, I am sure they are saving significant money on their chips.

You can spec the 14” & 16” to well past $4k also.

a Ferrari sports car is very 'expensive' and not many ppl can afford it, a Ferrari at 35% discount is a great deal but still 'expensive' for most ppl.

Bringing this back to the M1 Macs, I think what you get for your money in terms of computing, power consumption and build quality is simply amazing ! I think few can argue that. But in terms of 'absolute money', I def won't be able to afford a new M1 not when a "decent-workhorse" laptop cost about $900-$1400.

Not saying I don't want one ! I absolutely do,but my broke ass can't afford one :/

Before the M1 Air an entry-level MBP was the only "serious" choice for productivity. Even if you weren't a professional, an Air could be limiting.

Now we have an $800 laptop that can compete with the previous MBPs. The MBP can actually be a pro-only machine.

I guess also, are the 4 unused cores always the same?

As far as which cores are demoted… I definitely don’t know enough to speculate. If favoring the same set would affect longevity of the device, they’d rotate. If not it’s simpler not to do that. I have no idea which is more likely.

“I suspect that Apple has done this to further improve energy efficiency and ensure good responsiveness to new CPU-intensive tasks.”

Probably the "high power" mode available on an M1 Max in a 16" chassis is doing just that. And maybe Apple will decide to treat some kind of new iMac Pro as more of a pro machine, and make it chunky enough to dissipate that sort of heat without needing much fan noise...

Obviously a 'Pro' version of the iMac is coming eventually with a larger screen but I don't think many people expected that yet. Maybe February or so next year...

Also, even a huge company like Apple would want to stagger releases so design teams can have some constant workload. Releasing laptops and a new Mini and a new iMac would put quite some work on the team and then it would be idle for a time. They are probably done with the Mini and working on a Mac Pro right now.

My 15.4" can scale to "looks like 1920x1200" and the 14" to "1800x1169" so there's not a lot of difference. Does the smaller size make it feel cramped?

That the 16" also has more battery and better cooling don't hurt any, but they're the cherry on top.

That would equate to less "real estate" than what I have on my T460s FHD screen? Of course the screen would be light years ahead to look at. But just sounds a little bit strange with regards to actual pixels/space on the screen?

Yes, Apple doesn't provide for running at the native resolution out of the box (utilities like SwitchResX make those options available, though I don't know if they work with the M1s yet), all the display modes they propose are scaled.

Historically there was a default mode at 2x scaling with 2 above and 2 below, apparently on the M1s the default 2x scaling is the second-to-last and you only have one mode above.

So on the 14" with a physical 3024 x 1964 display, the modes provided natively by macOS are:

    1800 x 1169 (1.68x)
    1512 x 982 (2x, default)
    1352 x 878 (2.24x)
    1147 x 745 (2.64x)
    1024 x 665 (2.95x)

Don't skimp on RAM or buy into the hype that ARM halves the amount of required RAM - it's total BS.

https://github.com/srijs/rust-crc32fast/pull/6

(Get rust nightly via https://rustup.rs, pull the repo, run: cargo bench --features nightly)

    test result: ok. 0 passed; 0 failed; 4 ignored; 0 measured; 0 filtered out; finished in 0.00s

     Running unittests (target/release/deps/bench-b9bcbb53a5ef2c28)

    running 4 tests
    test bench_kilobyte_baseline    ... bench:         310 ns/iter (+/- 70) = 3303 MB/s
    test bench_kilobyte_specialized ... bench:          48 ns/iter (+/- 5) = 21333 MB/s
    test bench_megabyte_baseline    ... bench:     298,273 ns/iter (+/- 42,157) = 3515 MB/s
    test bench_megabyte_specialized ... bench:      45,100 ns/iter (+/- 7,249) = 23250 MB/s

    test result: ok. 0 passed; 0 failed; 0 ignored; 4 measured

(14-inch, 2021) - Apple M1 Max - 64gb

    test result: ok. 0 passed; 0 failed; 4 ignored; 0 measured; 0 filtered out; finished in 0.00s

     Running unittests (target/release/deps/bench-3eea1c397faf4328)

    running 4 tests
    test bench_kilobyte_baseline    ... bench:         223 ns/iter (+/- 8) = 4591 MB/s
    test bench_kilobyte_specialized ... bench:         100 ns/iter (+/- 0) = 10240 MB/s
    test bench_megabyte_baseline    ... bench:     231,689 ns/iter (+/- 1,885) = 4525 MB/s
    test bench_megabyte_specialized ... bench:     122,382 ns/iter (+/- 4,651) = 8568 MB/s

    test result: ok. 0 passed; 0 failed; 0 ignored; 4 measured

And the 5nm M1 has ~2.5Gflops/W which is not a huge increase compared to the 28nm Pi 4 at 2Gflops/W.

No-moores law in effect. Game Over!

Just want something that is more "apples to apples" (no pun intended but yes... lol)

Edit: Ok that is some sort of SIMD/GPU... I need figures for the raw CPU that can also do general purpose calculations and branches!

Maybe MIPS is a better "apples to apples" measurement?

I just ran it on CPU, single core (no pinning), just using the aarch64 fmla instruction.

Using pure BLAS (multithreaded, with the MM ISA extension), I get like 1.2TFlops in fp32 on the whole device. Code for the blas test: https://jott.live/code/blas_test.cc

And the 5nm M1 has ~2.5Gflops/W which is not a huge increase compared to the 28nm Pi 4 at 2Gflops/W.

No-moores law in effect. Game Over!

Same as these tests: http://web.eece.maine.edu/~vweaver/group/green_machines.html