Apple announces new Mac Mini, Air + 13" MBP featuring their own M1 chip.

[Chapelle]

---

[Passing Bye]

Btw, first singlethread results for Logic are out on GS (I created a litle test project and someone with an M1 MB Air has been so kind to run it). I used one EXS instance and ran it through an array of Space Designers (all serially on auxes running into each other). Space Designer is Logics IR loader also offering a socalled syntesized IR (which I think is some white noise that just gets some tweaking). CPU hit is in the medium range so it's perfect for comparisons.

Ok, on my 2x2.66GHz cheesegrater Mac Pro (still a pretty decent computer), at 44.1 and 64 samples, I can run 24 instances of Space Designer (quite good for a 10 year old computer). Someone with a 16 core new Mac Pro managed 52 at those settings.

And now fasten your seatbelts: The baseline Macbook Air (yeah, with just 8GB of RAM, even if that shouldn't matter much for this test) pulls of 80 (in words: EIGHTY!) instances! A fanless Starbucks gadget is running circles around a 12k machine in terms of single core performance!

Good stuff, really not surprised anymore, my partner filled me in with video side and seems that Air M1 is killing few k's Mac Pro's for video editing in practice, crazy stuff.

[Sascha Franck]

Serial signal flow doesn't imply single threaded or single core processing.
Btw it's serial only on summing say if you have master. Lot of people have multichannel analog summing so it's parallel.

Seriously, did you even read a single word of my explanation several posts ago? Apparently not so. Hence, I won't be debating this issue with you any longer, doesn't make any sense at all.

[Sascha Franck]

No way he could run 200 or 300 instances with 8GB RAM.

Completely depends on the kind of Alchemy patch.
Apart from that, I think the entire way RAM is adressed is completely different with those M1 chips.

[Chapelle]

---

[agharta]

I think the entire way RAM is adressed is completely different with those M1 chips.

Only for apps that are sharing a lot of data between different engines on the chip, due to the Unified memory architecture.
For apps that predominantly use only the CPU there's no difference.
The memory being on the same package as the SoC doesn't give a speed advantage.
Based on the data I've seen it has the same bandwidth as the current Intel mobile chip which has the RAM soldered on the system board.

[teilo]

Thanks very much - so, the message is that the Mini is really very good for the price, but some of the benchmarking showing it as better than any other Mac was a bit too good to be true when it comes to DAWs.

iMac 27 2020 i9 3,6Ghz 10-core reaches 317 Alchemy tracks.

Mac Mini 2020 M1 8-core reaches 126 Alchemy tracks.

I'm close to tears. Straight disappointment this real-world comparison.

You are forgetting that only 4 of the cores are high-performance (Firestorm) cores. The other 4 are high-efficiency (Icestorm) cores. It's very likely that those 126 tracks are only running on the Firestorm cores. According to that math, if there were 10 Firestorm cores, that would be 315 Alchemy tracks. This means it's performing nearly the same, but at a fraction of the power.

Once Apple scales up this architecture to pro-level, Intel is done.

[7blanche]

Thanks very much - so, the message is that the Mini is really very good for the price, but some of the benchmarking showing it as better than any other Mac was a bit too good to be true when it comes to DAWs.

iMac 27 2020 i9 3,6Ghz 10-core reaches 317 Alchemy tracks.

Mac Mini 2020 M1 8-core reaches 126 Alchemy tracks.

I'm close to tears. Straight disappointment this real-world comparison.

I haven't watched the whole video, but that test is flawed if you ask me. He seems to be using the 8GB RAM model of the Mac Mini. With that many instances of Alchemy, he must have run out of RAM. No way he could run 200 or 300 instances with 8GB RAM.

There is also Diva plugin test in that video:

24 for Mac Mini M1
68 for iMac i9

[Etienne1973]

It's very likely that those 126 tracks are only running on the Firestorm cores.

CPU display shows 8 bars rising together. Are all threads running at Icestorm speed? Otherwise meter doesn't reflect what happens or M1 has some x86 hyper-threading.

[Sascha Franck]

There is also Diva plugin test in that video:

24 for Mac Mini M1
68 for iMac i9

Diva is not yet natively ARM-ed but running in Rosetta mode.

[Etienne1973]

[Sascha Franck]

Well, that was to be expected. These days are over once and forever.

[spacepluk]

I did some testing in Logic using the suggested u-he chain (for science ).

And I'm seeing this both in realtime/playback:

Screen Shot 2020-11-19 at 10.21.57.png

Notice how Logic's meter shows usage in only one thread but the system's meter shows more. Enabling multicore for live tracks didn't have any effect. Logic has around 30 threads running and it was the only opened app. If I duplicate the track Logic shows it's using more threads.

I know this isn't Cubase but my point is that the system meter doesn't really tell us what we want to know. I still don't think it is possible to process serial signal paths in parallel, even when playing back recorded midi performances.

For most real-life cases I would still want a cpu with a lot of cores but strong single-core performance is really nice to have. I'll probably wait for M2 to give me both, hehe.

[ericj23]

Doesn't Logic background render tracks so that the current track is the only one that is "live"? Unless this has been turned off these test are not measuring computational power for processing VST but ability to stream audio files. While that is a useful metric it isn't really the one I want to know about!

Try the same test in Bitwig (or Studio 1, i think) as it does not have this background render function

[spacepluk]

Doesn't Logic background render tracks so that the current track is the only one that is "live"? Unless this has been turned off these test are not measuring computational power for processing VST but ability to stream audio files. While that is a useful metric it isn't really the one I want to know about!

Try the same test in Bitwig (or Studio 1, i think) as it does not have this background render function

I believe it uses a bigger buffer size for background tracks but it doesn't render. At least in this test it's not rendering otherwise I wouldn't see more than one bar when I duplicate the track.