Apples new M1 microprocessor

[olkipukki]

As per below test, M1 performs quite well taking into account these apps not yet optimized!

https://www.pugetsystems.com/labs/articles/Apple-M1-MacBook-vs-PC-Desktop-Workstation-for-Adobe-Creative-Cloud-1975/

P.S.
Not sure why they compared high-end Ryzens and didn't include Intel macbooks into the comparison...

[bd139]

Probably because it’s embarrassing.

My M1 mini is out for delivery today. Decided to avoid apple keyboards this time so grabbed some mac key caps for my existing one.

[Berni]

Well Pudget Systems is in the business of selling PCs so they probably wanted to make there prebuilt PC look good, so they used latest gen Ryzen since those are the fastest x86 out there right now.

But they do also point out in the article that both Macs perform about the same as a Intel Core i5 8600 or Intel Core i7 6700k https://www.pugetsystems.com/benchmarks/view.php?id=16278

None the less these are comparisons to desktop CPUs that have massive TDPs at there disposal so its not quite a completely fair comparison.

[Zucca]

My M1 mini

does it run FreeBSD on it? 

[bd139]

Not yet 

[DiTBho]

Personally, I would like to see and use Haiku (~BeOS) on a Bookair/M1! It's my secret dream

[bd139]

It arrived. Not going to get to play with it for a couple of hours yet 

[Cerebus]

It arrived. Not going to get to play with it for a couple of hours yet 

020 8744 2211 and damn balanced nutrition for the kids.

[bd139]

Hahaha. In preparation I already had a kebab for lunch and they're self serving oven pizzas 

Problem is a conference call which I'm still on now 

[Mr. Scram]

Well Pudget Systems is in the business of selling PCs so they probably wanted to make there prebuilt PC look good, so they used latest gen Ryzen since those are the fastest x86 out there right now.

But they do also point out in the article that both Macs perform about the same as a Intel Core i5 8600 or Intel Core i7 6700k https://www.pugetsystems.com/benchmarks/view.php?id=16278

None the less these are comparisons to desktop CPUs that have massive TDPs at there disposal so its not quite a completely fair comparison.

The same could be said for comparing the tightly knit Apple ecosystem and technology to more generic computers. By controlling both hardware and software you can optimise and integrate a lot more. It's never going to be a fair comparison either way.

[Mr. Scram]

Hahaha. In preparation I already had a kebab for lunch and they're self serving oven pizzas 

Problem is a conference call which I'm still on now 

I thought you of all people would have a loopable sound file with roughly appropriate ahs and ums.

[bd139]

Well that call overran until nearly 7pm. And it was mostly complaining loudly about a gigantic shit show!

I'm using it now and it's notably faster than the i3 Mac mini I had before, which I expect. Also it's always cold to the touch unlike the i3 which was actually uncomfortable if you annoyed it. I haven't done any benchmarks or compilation yet. Mostly been arguing with keyboard setup, turning all the shit bits of macOS off and getting the important bits working (Apple Music etc).

So far so good. It's nice to be using a Mac with proper keyboard though. The Mac chiclet keyboards are quite frankly dire things.

I'll do some more objective tests and comparisons to my other desktop over the next few days. I have a couple of mega sized C# things to throw at them both and it looks like there is CoreCLR runtime preview for native machine architecture is sort of working.

It's currently installing Xcode / compiler tools.

[brucehoult]

My M1 mini

does it run FreeBSD on it? 

In a VM, sure.

Apple seems to be saying it's ok to run other OSes on them -- they said "it's up to Microsoft" whether Windows will run on it -- but at the same time there doesn't seem to be documentation of how to do it and no one has figured it out independently.

On the other hand Apple provides as standard a hypervisor framework (low level -- you just get a memory range and EL0 and EL0 privs for code, and handle EL2 calls in your driver loop) and a Virtual Machine framework (high level, with virtio already implemented for you).

[bd139]

Ok a few hours in now and the UI latency is definitely down compared to any Intel Mac I've used. Their assertion is that this is due to higher performance of retain/release operations in their reference counter. I can believe that. I think so far that the majority of the observable rather than measurable performance improvements may come from that.

"zippy" is the word I'm giving it now. Also it has been running ffmpeg for 10 minutes (I haven't compared it to my Ryzen on this file yet) and it is still freezing cold to the touch.

[borjam]

In a VM, sure.

Apple seems to be saying it's ok to run other OSes on them -- they said "it's up to Microsoft" whether Windows will run on it -- but at the same time there doesn't seem to be documentation of how to do it and no one has figured it out independently.

Documentation, that long lost art!

I guess they will still be open sourcing Darwin.

[bd139]

Ok so first Rosetta 2 application installed this morning: MS Remote Desktop. No issues. Seems extremely quick.

[Marco]

Ryzen 4800U is currently the fairest x86 to compare against and that is a node behind ...

Apple's huge volume and margins make it kinda hard for anyone to compete for fab space.

[bd139]

Ok just signed out. Did a full day of work on it. Mostly RDP / zoom

So far no hard figures but some points:

1. I noticed absolutely no lag or performance issues. It's smooth as butter.
2. Runs ice cold. I just stuffed 75 gig of stuff into Photos app. On the old i3 minis they got burning hot doing this. This isn't even warm to the touch.
3. Runs silent. Despite being actively cooled it's not audible at all even under heavy load.
4. Zoom and MS RDP client were used most of the day, both x86 so Rosetta dependent. No issues.
5. Idle power is 5W. Peak under load was 20W (!)

I've got mine plugged into a 27" 4k screen so it feels like you're using an i9 + SSD equipped 27" iMac with a crap ton of RAM in it. But it's literally bottom end hardware.

xcode + toolchain currently installing.

The only bad bit I have found is the HDMI connector is a bit dicky. If you nudge it, it drops out. I've used two different cables and it is repeatable across both. That's being replaced with a USB-C to displayport cable tomorrow when it turns up so I'm not that bothered about it. HDMI has always been complete ballsack of a standard and connectors.

Edit: plus it's rather nice looking, better than the old space grey ones.

[nfmax]

connector is a bit dicky. If you nudge it, it drops out. I've used two different cables and it is repeatable across both.

I would make a warranty claim to Apple about it. That way:

• You get a new computer with a connector that won't drive you crazy each time you forget why you weren't using it
• Apple are made aware of a build quality issue
• Your M1 mini will eventually make its way into the refurb store, where cheapskates like me will be able to pick it up in few month's time

Tying this on a refurb 2018 Mini (3.2GHz 6-Core i7, 32GB RAM, 1TB SSD), which is actually not bad at all

[brucehoult]

Ryzen 4800U is currently the fairest x86 to compare against and that is a node behind ...

The 4800U has effectively twice the number of cores. The small cores on the M1 are worth about the same for performance as HT is worth on Intel/AMD. I felt bad enough pitting my 4500U machine against the M1 -- which won by 31% on the software build benchmark anyway without even using the small cores.

Apple's huge volume and margins make it kinda hard for anyone to compete for fab space.

This is of course not the customer's concern.

Benchmark sites show the M1 beating the 4500U on Cinebench R23 multi-core by 7760 to 5620, but losing to the 4800U's 10156.

From a customer's perspective, if you can buy a machine with a 4800U for the same price as a machine with an M1 and you mostly care about multi-threaded tasks and you are OS-agnostic then you should get the 4800U.  If your tasks are more single-threaded then the M1 is still faster.

[magic]

This chip is surprisingly decent and it looks like it could spell trouble for x86 if it can be scaled to higher TDP - crank up the clocks, double / quadruple the cores, etc.

You mentioned it's supposed to have peak throughput of 8 IPC. Are there any tools on OSX to measure it and see if it really makes a difference? From your experience designing CPUs / optimizing software / whatever you did (I suppose), where do you think may be the point of diminishing returns where going to a wider core just ceases making sense anymore?

[nfmax]

Perhaps the most interesting thing is that the new M1 Macs are cheaper than the corresponding Intel-based machines they replace, as well as outperforming them and being lower-powered. I suspect part of Apple's motivation was to avoid paying the 'Intel tax'.

Hopefully this will persuade Intel to up their game. They seem to have lapsed into the 'Soviet Weightlifter Syndrome' - breaking new (speed) records every year, but never by too much, as there is no percentage in that. Well there is now.

[bd139]

Yeah they are pretty cheap. I got discount and cash back on this mini so I paid £630 for it in the end. A comparable NUC is actually £70 more and not half as good.

[brucehoult]

You mentioned it's supposed to have peak throughput of 8 IPC. Are there any tools on OSX to measure it and see if it really makes a difference? From your experience designing CPUs / optimizing software / whatever you did (I suppose), where do you think may be the point of diminishing returns where going to a wider core just ceases making sense anymore?

8 IPC is not achievable on typical software, even with perfect branch predition.

I think it's useful on "inefficient" software such as that generated by JIT compilers, or languages which are doing bounds checking or reference count updating in parallel with the actual useful instructions.

[SilverSolder]

[...] I suspect part of Apple's motivation was to avoid paying the 'Intel tax'. [...]

As consumers, I guess we now have a wider choice... Intel tax, Apple tax, or AMD tax!