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

[Chapelle]

---

[david.beholder]

CineBench R20 vs. R23. You're comparing apples and oranges.

For example, the Ryzen 3990x scores around 24000 in R20 vs. 57000 in R23.

That's classic.

I also don't understand why everybody is so interested in single core performance.
Current OS is all about multicores and only one fast core is actually bottleneck.

[Double Tap]

CineBench R20 vs. R23. You're comparing apples and oranges.

For example, the Ryzen 3990x scores around 24000 in R20 vs. 57000 in R23.

That's classic.

I also don't understand why everybody is so interested in single core performance.
Current OS is all about multicores and only one fast core is actually bottleneck.

I could be wrong, but I thought that because audio is entirely serial, that makes single core performance more important than other applications.

[Passing Bye]

[Etienne1973]

I could be wrong, but I thought that because audio is entirely serial, that makes single core performance more important than other applications.

Meet some guys, form a band, go on tour, multitrack each night and have a good after show party.

[Sascha Franck]

I also don't understand why everybody is so interested in single core performance.
Current OS is all about multicores and only one fast core is actually bottleneck.

Well, in case you're wondering about that indeed, you should possibly not try to come up with all too much of those snarky and - uhm, shall we say "semi-intelligent" comments that we've seen from you so far in this thread.

But hey, I'll explain it to you anyway, being the nice person that I am.
For starters: Nobody's talking about "only one fast core". Quite obviously, once your arrangements become more complexed, you want multiple cores.
But still, especially when working with CPU hungry plugins including some plans to play them in realtime (or monitor your audio input through them, which is what you'd do, say, when using guitar amp plugins), a CPU delivering plenty of singlethread juice is needed because serial processes can't be distributed to multiple cores/threads. And up to this day, there's single plugins (not even chains of them) that will bring the most powerful machines pretty much to a halt once you use them in realtime. Just try a few of the Nebula plugins out and you'll know what I mean.
This is even more true once you're dealing with very low buffersizes, such as 32 samples - which is a pretty nice thing to be able to (in case you care about latency).
And it becomes even more important in case you work "modern dance style" alike. People doing techno, house, trap and what not often set up an almost full fledged mastering kinda signal chain on their master bus while composing already. And they're as well using plugins such as IKMM's Tape - which is a CPU hog of the finest sort. Quite obviously, along with the track you're just dealing with, these have to be processed by the very same CPU thread - simply because they're part of the serial chain.

All these things combined will easily show your CPU its limits.

Personally, I'm using a 2010 Mac Pro - and even if it's aging, it's still extremely powerful when it comes to multithreaded tasks. I can call up plugins almost endlessly and pretty much never max it out at all (at least not doing the things I usually want to do). Logic is adding quite a bit to that experience as by now, multithreading works quite excellent - plus, it's automatically switching the buffersize of all playback tracks to 1024 samples, which makes a *huge* difference compared to 32 samples.
But while all that is fine and dandy and would possibly suit me for the rest of my musician's life (apart from the fact that Apple is rendering my machine obsolete, but we should not discuss that right here), realtime singlethread performance leaves quite some things to be desired. I can not run everything I'd like to run - even less so at 32 samples buffersize. That's why I'm still using external solutions for my guitar amp needs, but I'd like to be able to work without them as well and use plugins instead (especially when not at my place(s) but on the road).

So, does that answer your question?

[jancivil]

CineBench R20 vs. R23. You're comparing apples and oranges.

For example, the Ryzen 3990x scores around 24000 in R20 vs. 57000 in R23.

That's classic.

I also don't understand why everybody is so interested in single core performance.
Current OS is all about multicores and only one fast core is actually bottleneck.

I could be wrong, but I thought that because audio is entirely serial, that makes single core performance more important than other applications.

I don't know what 'because audio is entirely serial' means. Distribution to cores isn't so much about how the audio is experienced or achieved live. You're right about certain applications but axiomatic that certainly isn't. A number of things happening at the same time but standing in line for the one shot at CPU time is not so viable for too many things.
There is no way I could have this project, or the last one, or... up at all [let alone performing well] on less than 4 physical cores, it would be too time-consuming involving a whole lot of freezing and premature rendering/revisiting.

So what isn't so relevant for me, and this is experiential not abstract, is very high clock speeds. IE: I have 8 [physical; 16 logical] cores on 2.3ghz far out-performing the last setup with 4 cores and 2.9ghz.

[Sascha Franck]

I don't know what 'because audio is entirely serial' means. Distribution to cores isn't so much about how the audio is experienced or achieved live. You're right about certain applications but axiomatic that certainly isn't. A number of things happening at the same time but standing in line for the one shot at CPU time is not so viable for too many things.

But that is precisely what's happening in the scenario I described above.
Want to process your synth's output with a chorus? Can't be done in parallel, your signal will have to finish the synth's processing before it gets to the chorus.
The very same is true for the internal architecture of the synth. A voice has to be generated before it gets sent into a filter. Again, this can't be done in parallel.
And so it goes on and on.

Sure, you could possibly jump between CPUs, processing the synth on one core and then processing the chorus on another core, but it'd still be a serial process, given the audio signal flow (which has to be maintained the very same way as in the analog world - plain logic dictates that).

Now, stacking CPUs serially *is* possible in computer land, I can take advance of that on my Line 6 Helix (it's got two signal paths, each using a dedicated CPU, which you can connect serially). But that will add one full cycle of processing latency. On the Helix, a dedicated hardware device, that''s bearable as the entire processing latency is something around 1ms (or maybe 1.5ms), given that the entire device latency comes at roughly 2ms (processing latency plus converter latency), but even on most modern computers, that's usually not a realistic value. Besides, apart from adding processing latency, it'll likely make latency compensation a way tougher thing to deal with.

Fwiw, I'm have no idea about programming, so there's possibly something I'm getting wrong, but as said, most of this is dictated by rather simple logic - you simply can't process distortion and delay parallely in case you want your distorted signal to run through a delay. Not even the fanciest code will get around that.

[Etienne1973]

... but it'd still be a serial process, given the audio signal flow (which has to be maintained the very same way as in the analog world - plain logic dictates that).

All events happen serial in the universe. It doesn't even matter if time is linear. Everything follows the law of the finest time resolution - theoretically. Who says there's a limit?

[david.beholder]

I also don't understand why everybody is so interested in single core performance.
Current OS is all about multicores and only one fast core is actually bottleneck.

I could be wrong, but I thought that because audio is entirely serial, that makes single core performance more important than other applications.

It's not. Because all DAWs are multicore. Remember first gen Ryzens were great at video and bad in audio at low latencies?
It's easy to create tests that would work great on certain processors (and vise versa) and not so great in others because of specifics like say pipelining that's why synthetic single core tests are quite meaningless.

[david.beholder]

Want to process your synth's output with a chorus? Can't be done in parallel, your signal will have to finish the synth's processing before it gets to the chorus.

That's not true anymore even in case of VSTs. Diva is working perfectly in multicore mode i.e. rendering voices on different cores, summing them and then applying chorus with only 16? samples latency - i.e. better than Ableton compressor. If you have more then one plugin chain/channel in Ableton they are (most likely) working on different cores and then summed to master.

[Sascha Franck]

If you have more then one plugin chain/channel in Ableton they are (most likely) working on different cores and then summed to master.

Not in case there's a serial audio signal flow.
Besides, Diva is an exception and I think even U-He recommends to switch multithreading off in case your host is doing fine with multithreading.
Apart from that, this will only work with synths anyway, as you have discrete voices that can be processed parallely. Anything serial will have to be processed serially - there's no way you could calculate a delay at the same time as an overdrive on a separate core in case you plan to have your overdriven sound delayed.

[Markus Krause]

On modern hosts every synth-plugin usually runs on a separate core. If you got more VSTis than cores the load is distributed amongst the cores.
There are also a few plugins that support multicore. Icarus2 for example makes use of 3 cores per instance, RayBlaster uses 4 per instance.
That's why it makes sense to lots of cores when you run plugins

[Sascha Franck]

On modern hosts every synth-plugin usually runs on a separate core. If you got more VSTis than cores the load is distributed amongst the cores.
There are also a few plugins that support multicore. Icarus2 for example makes use of 3 cores per instance, RayBlaster uses 4 per instance.
That's why it makes sense to lots of cores when you run plugins

Still, serial signal flow is serial signal flow.

[re.mark]

Want to process your synth's output with a chorus? Can't be done in parallel, your signal will have to finish the synth's processing before it gets to the chorus.

That's not true anymore even in case of VSTs. Diva is working perfectly in multicore mode i.e. rendering voices on different cores, summing them and then applying chorus with only 16? samples latency - i.e. better than Ableton compressor. If you have more then one plugin chain/channel in Ableton they are (most likely) working on different cores and then summed to master.

Ableton probably isn't the best example to use here when talking about multicore as it assigns threads based on track routing. 10 channels all grouped to one buss, that's one tread. 10 more channels and their group buss is sidechained to group 1, that's assigned to the same thread as the first. So your routing can do you over if your not careful.

Anyway, eager to read some production based reviews and test results soon rather than just benchmarks.