Apple Silicon 200 GB/s vs 400 GB/s Memory Bandwidth

[MeanMrMustard]

I'm hearing a lot of conflicting opinions on how Memory Bandwidth affects DAW performance. Apple gives us two choices of either 200 GB/s or 400 GB/s with the Max or Pro.

Does this actually make any difference in the audio/DAW world or is the higher speed memory bandwidth mostly for video/graphics users ?

MMM

[FlareStar]

No difference at all. Even with Mac Mini M1 68.25GB/s vs 400 GB/s.
You'll only see some difference rendering huge 8k footages.

[xsorter]

No difference at all. Even with Mac Mini M1 68.25GB/s vs 400 GB/s.
You'll only see some difference rendering huge 8k footages.

Plus this. It's purely marketing numbers

[pdxindy]

I'm hearing a lot of conflicting opinions on how Memory Bandwidth affects DAW performance. Apple gives us two choices of either 200 GB/s or 400 GB/s with the Max or Pro.

Does this actually make any difference in the audio/DAW world or is the higher speed memory bandwidth mostly for video/graphics users ?

MMM

Now it is 800 Gb/s with the M1 Ultra

[superscan]

No difference at all. Even with Mac Mini M1 68.25GB/s vs 400 GB/s.
You'll only see some difference rendering huge 8k footages.

Plus this. It's purely marketing numbers

I'd be curious to see the impact with large Kontakt libraries or instruments with samples. I'd expect those users to see appreciable gains, but most may not.

[Synth Master Jedi]

Has this been confirmed that there is no difference at all when using a DAW?

If a person is sitting on a Mac Mini or a Mac Studio Ultra, the extra memory bandwidth is not going to make a difference at all in their DAW session?

There's no difference between 68.25GB/S VS 800GB/S in real world DAW usage?

[Super Piano Hater 64]

There's no difference between 68.25GB/S VS 800GB/S in real world DAW usage?

Let’s do some quick math here. If your project consumes 70 GB/s memory bandwidth when played back in real time, that’s because every second of audio consumes that much RAM or more. Now suppose all that data is audio tracks instead of sampled instruments, and let’s say the whole thing is sixty seconds long. At a rate of 70 GB/s, the size of your project is thus in the neighborhood of 4.2 TB. If that sounds like the kind of scale of what you’re doing, go ahead and worry about memory bandwidth. If not, don’t.

[Synth Master Jedi]

There's no difference between 68.25GB/S VS 800GB/S in real world DAW usage?

Let’s do some quick math here. If your project consumes 70 GB/s memory bandwidth when played back in real time, that’s because every second of audio consumes that much RAM or more. Now suppose all that data is audio tracks instead of sampled instruments, and let’s say the whole thing is sixty seconds long. At a rate of 70 GB/s, the size of your project is thus in the neighborhood of 4.2 TB. If that sounds like the kind of scale of what you’re doing, go ahead and worry about memory bandwidth. If not, don’t.

Do you know how to check how much bandwidth a project is using?

If I open up one of my Logic sessions, is there a figure I can see somewhere which tells me how much bandwidth it is using?

I use a combination of audio tracks, sampled tracks and virtual instrument tracks in my projects.

[FapFilter]

There won’t be any differences at all for DAW use.
Maybe, just maybe for high end real time graphics performances at very high refresh rates, since CPU and GPU have to share the same memory

[PAK]

It depends on the app, which isn't really a great answer. So think about it like this..

Memory bandwidth matters most when something needs to constantly copy a whole bunch of data into memory. Like, say, a computer game. It might load huge textures for your 4k display, apply calculations to transform them, then output etc. This is why, if you look at Nvidia's graphics cards specs, you'll also see high memory bandwidth numbers being mentioned (Some 3090's might even go above 800GB/s!)

So, for Apple Silicon, it's mostly there for the graphics part. But the CPU gets the benefits too. There's a number of 3D / graphics / encoding things which can benefit. For audio apps it's mostly not the case. Let's say you have a 192kHz / 64 bit audio file. That's about 3MB/s of data. So even 1,000 tracks is still only at 3GB/s! In theory you could generate some higher numbers from that, depending on what a plugin does, but you kinda get the picture.

Mostly you just care about having enough memory bandwidth to keep all of the CPU cores fed with data which, in turn, hopefully stops stuff like audio glitches from happening! So it's not that it's not helpful.. your computer also has to do other stuff, and a 4K screen generates a big chunk of data even if it's just a DAW running on it. It's just that the numbers are overkill for many audio scenarios.

Apple are targeting video professionals with these chips, and you can't blame them - Just like the guitar market is bigger than the synth market, so the video production market is bigger than audio production alone. But, ideally, they'd offer an option with less graphics cores, and more CPU performance cores, for audio users.

BTW, and someone correct me if I'm wrong, afaik each core can "only" access 200GB/S. So it needs apps written to spread the load across the cores and get to 800GB/s. This is one area where M2 is likely to improve, upping the single core memory bandwidth amount further. But even the present 200GB/s is still wayyyyy faster than the fastest motherboard DDR5 PC memory just now (plus that DDR 5 is also expensive and can be difficult to obtain presently.. )

Outside of practical value for money considerations, you can't really go too far wrong with these Apple products. The one big caveat is if this size of purchase is going to have to last you a decade (or more) I would think hard about waiting for M2 later in the year. If the single core clocks also increase that's really going to help with audio performance. I'd hate someone to be in a position where they dropped a big chunk of change for a lot of cores, with lower speeds, when a cheaper solution with less cores, but at higher speeds, might've served them better..

[Super Piano Hater 64]

Do you know how to check how much bandwidth a project is using?

If I open up one of my Logic sessions, is there a figure I can see somewhere which tells me how much bandwidth it is using?

I use a combination of audio tracks, sampled tracks and virtual instrument tracks in my projects.

I'm not aware of any way for Logic to measure this information. The best we can do might be to estimate it, so let's continue with the thought experiment and see if we can get some more useful numbers.

Suppose every track in your project hits the worst-case scenario: It has to read a buffer full of samples, process them, and write them to another buffer before they can go to the mixer, which involves reading buffers for all tracks and writing them to the output buffers. (I'm going to ignore the processing overhead, and thus the difference between (for example) ESM and Diva, because CPU speed is not the bottleneck we're concerned with here.)

Let's say all tracks are stereo and you're working at a 192khz sample rate, the highest available in Logic (at least with my setup). I think it always mixes in 32-bit float. Let's also go with a buffer size of 1024 samples. Thus, every buffer read/write is 32768 bytes. If every plugin in your effect chain has to read and write two complete buffers for every processing step, that's 6,144,000 bytes per second per channel per plugin. So, the number of plugins you can be using simultaneously caps out at... about 5,500, or 55 plugins per track (including instruments) in a project with 100 tracks. Something like that.

How often do you work with projects of that scale?

(Looks like someone came up with different numbers while I was writing this. No big deal.)

[PAK]

(Looks like someone came up with different numbers while I was writing this. No big deal.)

Hehe, yep, trying to make the same point anyway Even for giant projects the amount of memory bandwidth on offer isn't an issue.

The real worry is if your chosen DAW can make good use of (especially the bigger) M1's available resources.

[Synth Master Jedi]

So, the number of plugins you can be using simultaneously caps out at... about 5,500, or 55 plugins per track (including instruments) in a project with 100 tracks. Something like that.

I do sometimes use over 100+ tracks, but I guess that there is nothing to worry about at all, since I will never use 55 plugins per track. I don't even use 5 plugins per track and I often use 0 plugins per track if nothing is needed.

Based on what has been said about the memory bandwidth, I think it will be easier to max out the CPU than the memory bandwidth in a DAW, if somebody is using very intensive plugins, like Acoustica stuff that is known for being extremely CPU intensive.

[Synth Master Jedi]

The real worry is if your chosen DAW can make good use of (especially the bigger) M1's available resources.

I'm assuming that Logic should be a safe bet, since it's made by Apple afterall.

I want Logic to use every available core, every available everything on an Apple Studio machine, so that somebody can get max performance when they are making music.

[Super Piano Hater 64]

Based on what has been said about the memory bandwidth, I think it will be easier to max out the CPU than the memory bandwidth in a DAW, if somebody is using very intensive plugins, like Acoustica stuff that is known for being extremely CPU intensive.

I tend to agree. Higher core counts seem like the best way forward for a beefy DAW setup. I don't have an M1 machine; I'll be keeping an eye on what they do with the M2 before I consider buying in.