Hacker News Reader

Comment by Aldipower | Hacker News Reader

I own 4 Mega STe and I am using them almost daily. One of them, the rest is spare parts. Producing music with it. The Atari is my MIDI master clock and central piece of MIDI sequencing together with Cubase 3.1 for the Atari. Seriously the MIDI timing is unbeaten until today! The MIDI ports are directly attached to the CIA chip which is again directly connected to the Motorola 68k CPU. Runs absolutely stable even 35 years later. No crashes what so ever and also no distractions by updates or "phone home applications". It just works, distractless! Shame on the "present future".

View on HN

> Seriously the MIDI timing is unbeaten until today!

Is this in any way related to the general "speed is going up but latency is getting worse" phenomenon of hardware in the last decades?

Yes, I think this is a on point statement. :-)

Yes, back in the days, I/O was often really low latency because memory and therefore buffers were expensive and gate count was limited, it meant more direct connections, which meant low latency.

The Atari 2600 for instance was known for "racing the beam", updating the image while it was being drawn on the CRT monitor. A latency measured in pixels rather than frames! It was necessary because the console didn't have enough memory for a framebuffer.

The Atari ST is special for the inclusion built-in of MIDI ports, and it was made cheaply, which at the time meant direct connections and it resulted it low latency.

You can have low latency and low jitter today, but you will need to use a microcontroller not a general-purpose CPU. The old 16/32 bit retro machines are essentially microcontroller architecture devices with general-purpose computer peripherals, for pretty much the reasons you mention. But there are many cheap microcontrollers available today, such as the Raspberry Pico series.

And when you factor in FPGAs, you can get down to the microsecond or less. Low latency is possible, it is just that priorities are often elsewhere.

We like being able to plug everything anywhere. And I admit it is damn cool being able to connect a display, most kinds of storage devices, keyboard and mouse, all while charging my laptop on a single port, at any time. I may even be able to disconnect my laptop and put my phone instead and it will do something sensible. If you did that back in the day, there was a good chance for one of the devices to turn into a smoke machine.

It comes at a cost though.

> If you did that back in the day, there was a good chance for one of the devices to turn into a smoke machine.

Back in the day, you would not have been able to do any of this with one port. Each type of device had it's own uniquely shaped connector/pins combo. You were not going to connect your SCSI devices into the VGA monitor port accidentally. Closest I ever saw was someone attempting plug in a Mac ADB cable to the S-Video port, but that just resulted in bent pins. It just so happened those pins were on an Avid Film Composer dongle instead of a replaceable cable.

I think modern general purpose CPUs are perfectly capable of low latency and jitter. The problem isn't the CPU, the problem is stuff around the CPU (mostly the operating system). The less deterministic aspects of modern CPUs (branch prediction, speculative execution, caches etc). happen at timescales much smaller than what you usually care about (and possibly smaller than the jitter specs on microcontrollers).

ps/2 keyboards trigger an interrupt. USB are polled.

A modern CPU polls faster than it can handle interrupts. That can be done on every nth iteration of the sound producing loop.

A rp2350 with psram and microsd could probably do a commendable job at pretending to be an entire Atari ST while providing a bootload of extra low latency goodies at the same time

I refer to the RP2xxx chips as "headless Amigas" because their PIO modules essentially function like Coppers: they are simple state machines that offload I/O functionality off the CPU.

I think there's a very strong future in emulation of achieving FPGA-like latency by using a Raspberry Pi Pico/Pico2 to emulate each of the target machine's subsystems/chips.

Have you seen https://github.com/floooh/chips A bunch of PIO linked chips using these interfaces would feel like a weird blend of software and hardware that stands separately to the FPGA world. I have wondered if it would actually work as a larger scale paradigm. Imagine a single piece of silicon with a bunch of RP2xxxx level processor/ram blocks with PIO links between them all. I'm not sure how it would come out compared to FPGAs for balance of flexibility/cost/power consumption/etc. but I suspect it could find a niche.

antirez mentioned running some of these on RP2040's

The 68000 is "essentially microcontroller architecture"? I don't think there are many people who understand architecture that would agree with that statement.

> The Atari 2600 for instance was known for "racing the beam", updating the image while it was being drawn on the CRT monitor. A latency measured in pixels rather than frames!

Oh wow! I remember hearing that oculus were doing this on their devices and thinking it was new.

This is true, but in my opinion also misleading. Speed and latency are fundamentally different. Speed would be a Performance Feature in the Kano model, meaning there is usually a linear relationship between speed and user satisfaction.

Latency would be a Basic Feature. Once you cross 7 ms (or 5 ms, or even 3 ms if you absolutely insist) you're happy, above that everything is absolutely unusable.

You are missing out the jitter. This is often the worst part of modern implementations. If there is a jitter of 4ms and peaking sometimes with 20ms, then a 5ms latency is still bad. This implementation is basically unusable. Like many modern USB ones..

The Atari has an absolute stable and extremely low jitter. Some guy measured it to 1µs. Cannot find the link though, sorry.

So the Atari has low latency around 2-4ms with an extremely low jitter. This is execatly what you want from a MIDI clock and sequencer driving multiple MIDI devices.

How do you think any professional works nowadays with MIDI? A good, modern USB interface (from Focusrite or similar) has a jitter well below 1ms, usually in the range of 200µs. If that is too much, simply sync your DAW with an external, dedicated clock, which will usually give you a jitter in the single µs range.

I have a Focusrite and the MIDI timing is terrible. Sure, there is more to it then just the interface. With USB you just cannot guarantee a stable midi timing, because there is no good midi buffer implementation for it. Technically it would be possible, but no one cares.. Professionals using something like MIDI to audio converters via an VSTi plugin that takes midi signals, modulates them onto a audio signal (which can be easily buffered) and some dedicated outboard equipment converts this back to MIDI. If you are working with hardware synths, etc. this is the only option you have nowadays with non-vintage hardware. A lot of producers do not work with midi anyways, they use plugins, that's why it is some kind of a niche problem and there's not much talking about it.

First off, I'm assuming of course we are talking Mac here, because Windows is unusable for MIDI. If you have terrible MIDI timing with a Mac, then yes indeed, you'll need to sync via audio, but there are nice and inexpensive solutions for this, for instance the Midronome.

Look, I'm not trying to convince you to get rid of your Ataris, quite the contrary. I'm just disagreeing that it's impossible to have low jitter nowadays, but I fully agree that things used to be simpler before everything was done via USB.

Agreed. It is of-course not impossible, but it is almost impossible out-of-the-box (literally ;-)) I have a USAMO (Universal Sample-Accurate MIDI Output) device, but do not use it, because as I said, Atari is king here. :-) Not sure how the Midronome can solve the problem of midi notes coming inaccurate from a modern DAW? But maybe I do not understand it completly. Need to have a deeper look. Since some years I am using Linux with a Focusrite for mastering and audio tracking. Midi was bad with Linux and Windows since I got my first USB interface and went away from PCI interfaces. But this shouldn't matter too much. :-)

> Not sure how the Midronome can solve the problem of midi notes coming inaccurate from a modern DAW?

Here's a review from a nice scotsman explaining how this works:

https://www.youtube.com/watch?v=XCZqkSH9peI

or a walkthrough from the creator:

https://www.youtube.com/watch?v=hkw9dmLfkZQ

Note that this is an old version, I just saw the there's now the "Nome II", and at least for Mac, he has actually developed a USB protocol to provide a stable clock (which as you've already written is totally possible via USB, it's just nobody cared enough):

https://midi.org/innovation-award/u-sync

For Windows, the sync is still done via audio through a special VST.

The YT channel by the creator has many more interesting stuff, he also has done very precise jitter measurements, see for instance:

https://www.youtube.com/watch?v=zGU336yKyEM

Thanks a lot! The scotsman is cool and his t-shirt too. :-D T-Shirt says in German "Little pig".

Regarding "midi notes" Sim'n Tonic himself is saying this to the Midronome: "Note that only these MIDI messages are simply forwarded when they are received, their timing is not changed. So if your DAW sends them with a lot of latency and/or jitter, the Midronome will forward them with the same latency/jitter. Actually this is a problem I plan on tackling as well [...]"

So the Midronome does not solve the problem of inaccurate midi notes coming from a modern DAW. The USAMO does by the way.. But only with one midi channel at once. And of course, coming back to the actual topic, the Atari hasn't a problem at all with accurate midi notes, it is absolutely tight at all 16 channels. So it seems there is indeed nothing comparable to the Atari nowadays. Maybe it will in the future.

Not sure if that is still accurate. This might only be available for Mac, but on the FAQ for Nome II it says this:

Can Nome II send MIDI Notes?

Nome II is like a MIDI hub, you can ask it to forward any MIDI sent over USB to one of its MIDI outputs. It will not only forward these instantly but merge them smartly with the MIDI Clock, without affecting it.

> because Windows is unusable for MIDI

This is simply not true. Many performers use Windows laptops and MIDI to control their stage equipment without issue.

The Windows MIDI/USB stack adds considerable amount of jitter to the MIDI clock, compared to the much superior ones in MacOS. I will fully admit that "unusable" is a personal opinion based on my experience. Of course performers also use Windows, but I heavily doubt you are able to see which device in their rack acts as a master clock, and how they sync their devices, apart from the fact that most performers nowadays don't use MIDI at all.

Midi is used heavily for guitar patch and lighting automation as well as triggering backing tracks in a DAW running on stage. The use of MIDI (over USB) has only increased on stages.

This is getting ridiculous, we are talking about making music, so triggering notes from different devices in sync. You know, what MIDI was originally designed for, not triggering some lights, guitar patches or a background track. You are exactly proving my point: MIDI nowadays is pretty much reduced to SysEx for doing simple automations. None of that is seriously affected by jitter in the ms range. You sound like you have no idea how electronic music was done before VSTs were a thing.

Not really.

MIDI is a serial protocol.

At any given time only one message can be sent down the wire. [1]

So on the beat, an implementation can send either the clock pulse or note on or something else. [2]

If you send the clock everything else has to wait. If you send something else, the clock has to wait.

Now with modern computers, you are also dealing with USB which is a low priority parallel protocol and has to coordinate with everything else a modern kernel does.

Music is hard.

[1] premium hardware sequencers sometimes have two or more Midi Out to reduce contention.

[2] Midi Time Code solves this by encoding monotonic time into Midi and is how serious sync is done over Midi, e.g. in Hollywoood

What's the S in USB?

FWIW you can get excellent latency on a modern device but only if you run everything in real mode and forgo complicated buses like USB that are effectively network link layers.

I really wish you could easily get somwthing like the Myster ST clones...seems like supply is spotty, and price seems pretty high. Id love an original if they were less marked up too...

> Seriously the MIDI timing is unbeaten until today!

I’ve got a full studio at home, but tbh i never know what people mean by this

I have the equivalent of a $500,000+ studio from my childhood, all in my laptop.

You are concerned about a 9600 baud protocol.

There is zero 'shame' on the 'present future' when it comes to music production tools. It is like one of the hugest bright spots/biggest equalizers. Best thing I did was go ITB. No headaches. No hardware maintenance on obscure hardware. No MIDI limitations or even considering of my MIDI chains. Just music making.

Some musicians still like to play instruments -- for them and their listeners, ITB production is seen as a cheat and not real musicianship -- and for them the lack of a stable MIDI clock on today's hardware absolutely does matter. A trained musician can feel time difference as small as 1 ms. Any latency or jitter greater than that and a perfect track could be ruined.

As an aside, all-digital workflows take the joy out of music being made in the moment, by ear and by feel. There is no replacement, for example, for a professional sound engineer adjusting a mix strictly by the sound in their headphones and the feel of the sliders under their fingers.

> A trained musician can feel time difference as small as 1 ms

No they cannot.

https://www.tandfonline.com/doi/pdf/10.1080/14992027.2025.24...

Great paper. So the average psychophysical gap threshold is at ~2ms! This is lower then I expected tbh, but always suspected.

You always want latency or jitter as low as possible. Latency adds up in the chain. Instrument/Sequencer (2ms) -> Some digital effect (3ms) -> Digital Mixer (3ms) -> In-Ear monitor bridges an air gap with processing (6ms) Suddenly you have 14ms latency. Bad, but reality. So, every ms less is better.

Regarding jitter, this is the worst, because the brain cannot adapt to the changes, whereas constant latency can be regulated somehow by the brain.

I have a Novation SL MkII as my controller keyboard. It is a much more tactile experience than say a DX7 or other menu diving synth. It has faders built in for mixing. As someone who has done both I have so much more joy being all digital. I have access to so much that I never did before.

There's not one word in his post where he looks down on VSTs or anything. It's just how he likes to make music, and he is unhappy with the state of modern MIDI implementations. In fact, it's the exact opposite: you are shaming him for still using MIDI.

His Shame on the "present future" comment disagrees with you.

Like C++, this cannot be parsed with a context-free grammar. The "present future" refers to

No crashes what so ever and also no distractions by updates or "phone home applications"

which is something I would guess most people would indeed see as shameful regarding our present future in software, but OTOH, this is HN, so who knows.

My old synths crashed, and required physical maintenance. In addition I lost songs that failed to read off of the crude tape backup that my Quantitizer/sequencer used.

I am much happier with my setup that I could have never have afforded outside the current future than my much lessor previous setup. This being HN I'm sure that are people that can afford to spend much more than me for gear so they might prioritize the 'minor differences' to them you list over no access at all, but I much prefer having access and price points I can actually afford.

Good for you! I also sold all my external gear many years ago and never looked back. And this is all very interesting, but has nothing to do with the topic at hand: where is he shaming people for using VSTs instead of external gear, or saying people are cheapskates for using VSTs, or that his music is better? Because he still doesn't. He says USB MIDI has tons of jitter and modern music software phones home and crashes a lot, and Cubase on his Atari does neither, which from my memory is totally true.

I have had zero issues because of jitter with my Novation SL MkII so I consider it a non-issue, and I think the tons of hits made using USB MIDI controllers speak to that. or the tons of artists that perform using Ableton. Or the tons of live artists that perform using Gig performer. Going between a piano, piano action, synth action, and un-weighted keyboards has much more impact on playing. Do we shame the old past for having different pricing options/cheapening out on keybeds?

There is so much music software that doesn't phone home it's a ridiculous focus. iLok is the biggest phone home and you can buy a USB stick to stop that so? I don't get the point?

" No crashes what so ever and also no distractions by updates or "phone home applications". It just works, distractless! Shame on the "present future".

Still doesn't make sense to me and is needlessly shaming/negative. My setup just works and is way more distract less than past hardware setup, especially if you are talking about recalling projects. I lost WAY more projects/work previously than I do now, and when things failed in the past you were hit until you could perform a hardware fix (if you could afford the cost of the fix). I'll take software.

MIDI baud rate is 31250, not 9600.

Reply on news.ycombinator.com