Hacker News Reader

Comment by smoldesu | Hacker News Reader

Counterpoint: M2 is a tock release, not a tick. This is pretty much the same behavior that ended up getting Intel shot in the foot: leaning too heavily on marginal performance increases from process enhancement will bite you in the butt when your competitors get their tech on your node. Intel and AMD are both going to be on 5nm in 2023, Apple should be pretty worried if this is what they're fighting with. 3nm isn't even getting taped off until 2024 AFAIK, so this really does feel like Apple's "Skylake" moment, as far as desktop CPU architecture is concerned.

View on HN

Spend a few minutes talking with someone who designs hardware. Often when they release something, there are TONs of incremental improvements that they just didn't have time to get to when a new product is released.

I remember having a conversation with someone who worked on hard drives in the 80s. He got so ^%$^$# excited telling me about all the improvements he worked on between generations; they were mostly things like tighter calibrations, and refinements.

Point being: Don't knock releases like this.

Indeed, that "18%" improvement is almost certainly made up of very many 1% improvements (or more likely, many improvements that each only makes difference in some scenarios).

that "18%" improvement is almost certainly made up of very many 1% improvements

My guess is 18 of them.

I'd guess 17, but I guess it depends on whether they compound. :)

Ah, the old synergy effect, a corporate executive's best friend.

This comment sounds like the exact same hopium I heard about Intel in 2016. I get what you're saying, and there are definitely smaller changes packed in here, but my point is that Apple's performance crown is looking mighty easy to usurp right about now. Hell, they even showed a graph with the i7 1280p beating the M2's single-core performance by ~20% at the WWDC today; they know they're on notice.

M2 isn’t intended to have high performance. It’s intended to have a high perf-power ratio, as they said.

Often it gets the performance too, but that’s kind of like a lucky break.

Still M1 Pro/Max/Ultra single core performance is basically same as M1. It's not good sign for M2 Pro/whatever's single core performance. It's fine for M2.

And why do I care about perf-power ratio if my laptop is plugged in? WHY? I can understand if it switched to high perf-power ration when Im not plugged, but why would I need that when Im plugged in? Seems like a made up, just because metric.

Because lots of power makes a system hot. And a hot system is a slow system, even if the battery is full.

https://www.novatech.co.uk/blog/cpu-and-thermal-throttling/

Nothing you said here doesn't not apply to M1 which also overheat under high load and gets throttled, and some of them dont even have a fan.

https://medium.com/illumination/how-to-solve-overheating-iss....

A lot of these benchmarks run for several minutes and test the system as it OVERHEATs, simulating real conditions and Intel i12 beats M1/M2. So this doesnt change anything or bring any new information.

This is the first sentence of that article, “MacBook rarely shows overheating issues.”

I can say that’s definitely not the case of the Intel laptop I’m currently using.

Otherwise it would throttle instantly because your MBAir has a passive cooling system.

Also, it would make you infertile and it’d drain the battery faster than the charger can recharge it.

That 20% better perf was at 4x the power though?

There's more to comparing computers than just CPU benchmarks. Benchmarks can be manipulated. Raw CPU only matters when your workload involves lots of CPU work. (And, if it does, maybe a desktop is a better choice.)

But, more importantly, it's like comparing two cars based on horsepower when in practice both will get you to the same place, on the same road, at the same time.

Apple doesn't really have to worry about competition too much for a long time, even if Intel and AMD get comparable watt/power levels for laptops, Apple isn't going to be using them for their products anytime soon. And people using Macbooks aren't likely to switch to Windows/Linux machines, even if there is a 20% better CPU.

But really we've only had 2.5 released cycles so far. Not much to go by.

I might be a very small minority but I'd personally switch back to Linux in an instant if there was something comparable to the M series hardware

I'm considering getting an M2 Air even though I very much don't like OSX.

Asahi linux has most things working, except the accelerated GPU. Just in the last week or so a triangle was rendered, so progress is has been made.

Sadly while apple has 128 bit wide memory on the low end (66-100GB/sec on the mac mini and mba), 256 bit wide (200GB/sec), 512 bits wide (400GB/sec), and 1024 bits wide (800GB/sec). Nothing wider than 128 bits looks to be coming to standard laptops or desktops in the non-apple world. I don't really count HEDT chips like the threadripper pro, since they are very expensive and very limited, and burn many 100s of watts.

Asahi Linux is, for me, the most interesting thing going on in Linux these days.

Agreed. I'd likely have a studio today if the GPU port worked, but unless it comes out RSN I'd likely wait for the M2 based studio to come out. The chip has other magic inside as well. I'm hoping Linux continues to implement support for the MatMul instruction (not just vector multiplies), 16 trillion op neural engine, various encode/decode video accelerators, etc. I've heard vague references to compressed swap to help make the most of limited ram (m1 limit was 16GB).

What package manager/distro will Asahi Linux be using/basing off of?

It's based on arch, but there are already people who ported the necessary changes to nix and something else I can't recall. The idea though is that once things get upstreamed, any distro will work.

Perfect, looking forward to when it's stable.

Massive memory bandwidth is for GPU, so not matter unless you utilize iGPU.

I disagree, Anandtech tested the M1 max (409GB/sec bandwidth), and the CPUs managed 224GB/sec with the faster cores, and a total of 243GB/sec with fast and slow cores.

Sure that's quite a bit less than 409GB, but there's other major parts of the chip like the GPU, video codec acceleration, and neural engine.

Seems pretty unique and useful to me, not everything fits in cache.

That's true that CPU core consumes massive available bandwidth on AT's test, but it doesn't mean that such bandwidth is used for real CPU usage. I don't oppose memory bandwidth is meaningful for total SoC including iGPU/ML/Video parts.

Asahi Linux is running well enough to be usable for lots of things.

However, no screen brightness control, no sound or YouTube currently. And there's some page size weirdness that means Chrome/Electron (?) is not usable, so no VS Code.

If you can live with that, it's so nice having Linux on the M1 hardware.

YouTube is the acid2 test of the modern web. I had a laptop that didn't hardware accelerate youtube, and it ended up as a headless server.

Despite the exciting news and I'll definitely upgrade my intel MBP to either M1 or M2 in the future, I doubt my personal workflow will change: Linux for development, Mac for productivity and maybe some (lightweight) development, Windows for gaming and gaming only. While Apple probably wants to steal existing user base from Windows, on productivity uses maybe, I don't think they're planning to drag people from Linux on development workloads.

M2 has a media engine, which was previously reserved for the higher end versions.

Considering that M1 was already an overkill for the most tasks, I think it is a meaningful "tok" upgrade.

Besides, I kind of expect the trend of ASIC to continue. Instead of having more extreme and extreme lithography, it kind of feels more appropriate to have computation specific developments.

AMD recently teased a ~15% single threaded performance increase for Zen4 when they will move from TSMC 7nm to TSMC 5nm.

Apple just teased ~18% CPU increase while staying on TSMC 5nm.

Sounds like they are doing just fine.

AMD stated >15%, not ~15%, and this was the worst-case, conservative number, as they have clarified.

The increased core frequency alone brings in around that number already.

https://overclock3d.net/news/cpu_mainboard/amd_calls_its_zen...

That 18% is multicore

My understanding is that AMD coasts behind Apple’s process progress at tsmc, seeing as Apple invests the bulk of resources into tsmc’s cutting edge nodes. If apple hits a wall, well so does everyone else depending on the fruits of their r and d.

Also, if Intel can keep up this time, and if tsmc/apple slouches, I wonder if there will be Apple silicon and Intel x86-64/risc macs/apppe devices on available simultaneously?

My understanding is that 3nm for Apple is coming in 2023.

https://www.slashgear.com/833760/apples-3nm-processor-is-abo...

Coming in 2022 for the iPhone 14 Pro line.

Only because it’s a COVID year.

Reply on news.ycombinator.com