Interesting. So you think I shouldn’t even bother with my Pi since it’s only got 1Gb? I can try to hunt down a 4Gb version but they’re getting harder and harder to find.
I’m stuck with the Pi3 because the cool little custom case is milled for it and we must keep the cool factor.
My though was with Hans review last week, I can just pirate the guide Magna Mano has for their streamer’s optimization using Gentoo because they also use a Pi3 B+ so I’d have a pre built roadmap to configuration.
Honestly I think it might not be. Also cool factor is vital and everything cool sounds better. 
Personally I think something like Ian Canada’s re-clocking and isolation hats or even a cheap DDC will have a far bigger impact than the specific nix OS as what we are really talking about here is clock speeds. Lower latency is a potentially faster clock but generally there are a rapidly diminishing gains to be found with OS tweaks since the onboard clock is still in a noisy environment.
Low latency kernels are aimed at music professionals for audio recording, allowing better sync between live and recorded signals. The point being we are aware of timing differences when we can hear two signals side by side. When we listen to music playback we do not do this. I am sceptical that a human can perceptibly hear a difference in millisecond clock speeds on a signal that passes along wires, through a DAC with its own clocks and then out to physical imperfect sound reproduction equipment.
I have played around with the kinds of settings and kernel options involved on both Raspian and Ubuntu and I couldn’t hear any difference at all. However all my sound has jitter correction/re-clocking/isolation post source so I don’t know if it can have a difference or if it is a difference that I personally am not sensitive to.
Note: If you have a system and ears and brain that allow you to hear a difference between different nix OS sound processes than that’s awesome! Enjoy your blessing (or curse) and more power to you! I personally, based on my knowledge and experience, think it unlikely to be a real difference but I am not here to harsh your vibe. If you subjectively hear it, you hear it and that’s cool, I’m not going to try to tell you to enjoy your music differently 
Interestingly, I’ve read in a number of places that the pico to femto second clock push is mostly for HW bragging rights and that in reality what you’re looking for in a clock is that it remains synchronized to that frequency through as wide an operating range as possible. That when you get a clock that can go down to Femto, that it doesn’t automatically mean that it’s better as a Pico clock is more than good enough to function at the Mhz range frequency if it remains consistent trough a broad window of operation. I’ve seen more and more vendors using four clocks (two for each frequency) and average them to get improved accuracy and achieving better results.
The whole oven temperature control clocks is to do precisely that, allow them to remain synchronized through a wider operating range.
And they still don’t solve the underlying issue with a none real time OS.
Linux is used in streamers because of the license free OSS, not because it’s a good choice for the job.
It’s just not feasible for most streamer manufacturers to do much more than customize a distribution and install some software.
There’s probably a quality win to be had messing with kernels, but it’s not likely to be huge.
Agreed. Unless a manufacturer is writing their own os for their hardware and is able to do a better job than existing general purpose OS creators it is always going to be a compromise. I think it safe to say that nix based is a better option than Microsoft based as it is easier to trim out fat, more stable and driverless but it will always still be a compromise. There is a reason dedicated pro audio equipment like RME sell for thousands and it is not the hardware, it’s the software development.
Different sort of clock though, the OS interrupt handling clock can only really deal with latency, not the output accuracy of the device, all the OS does is dump stuff into a buffer and hardware clocks it out on the audio clock.
Which is good because if you were relying on the OS clock to push samples out in a none RTOS, you’d have a lot of clicks and pops.
The thing that gets lost when we talk about clocks in audio is absolute time doesn’t really matter a clock running 1% fast or slow it far less relevant than the Jitter at whatever rate it’s running.
Here’s one for you, for devices that use a 10MHz external clock, it’s not exactly divisible by the 44100 sample frequency, so by using it as a timing source you introduce Jitter up to 1 clock or 10nS, which is huge in the world of pico Second clocks, or you live with a sample rate of 44052Hz. I do wonder which they do.
Note some manufacturers like DCS use an external word clock, which doesn’t suffer from this.
One of the big very highend server/streamer manufacturers is Windows based, Pink Faun I think.
Modern windows is possibly better for a Real TIme app, they added functionality to the Kernel circa Windows 8, to resolve some of the inherent latency issues.
But it’s certainly going to be easier for a manufacturer though, and any Windows based streamer is going to end up with a much beefier CPU, and more noise that needs to be suppressed.
Really interesting! I imagine it is again a question of investment. You can absolutely write good audio software to run on windows but it is much more complex and expensive than writing an interface on top of a nix OS allowing access to built in tweaks.
I remember the drama about the change to the Windows 8 sound stack. It messed up a lot of older music software! The biggest issue I have with it is how hard Windows makes it to avoid the built in equalization and how hard it is to route individual programs to different sound sinks without additional software, processing and general faffing about.
With Pipewire entering the mainstream at present as a replacement for both Pulse and Jack audio on linux is in a decent place at the moment.
Looking at the various versions the Gentooplayer has available for download, it’s pretty interesting to see that he’s got version specific to various CPU so he’s doing some optimization that’s not off the shelf.
I agree that probably the benefits are minimal and the hurdles higher. If I didn’t have to get a new Pi I’d probably try it. I still might try it even with just 1Gb of memory to see what it’s like.
I know is way off balance, but over the last few days I got a Shunyata Hydra AC conditioner and connected some decent power cables (most of which I had laying around) to my desktop and I’ve been floored by how much of a jump yet again the system has made just by addressing the basics.
Without getting too much in the deep end on expensive components, just investing in a good power and good cables, the ability of some of these lower priced offerings are REALLY good for the money especially considering what moving to the high end costs.
I’m wondering if we’ve (as a collective) been looking at the model all wrong and instead of upgrading components as you progress, you first upgrade the foundation AC power and a couple of good AC cables as well as good DC power supplies. Would we be more than happy enough as a collective with the mid-fi level gear. I know it’s a tough sell to someone first getting into the hobby to buy a good AC power conditioner for hundreds of dollars. But in retrospect it’s far easier going back now making a chain on the cheap and getting it 100% right with maximum synergy.
I know I’m just using headphones but a the digital front end is my custom Pi and a Denafrips Pontus II and it’s over the moon with the quality jump.
…anyway, just rambling now. It’s always fun to play with gear.
Most likely this is more to do with hardware age. Pi3 and 4 have different enough architecture to need differing versions of OS. The two versions for x86 and x64 chips look to be a headless and a version with a lightweight desktop renderer.
RTAI seems to have been doing an excellent job for real-time critical applications for years (decades?). Issue is running hard-real-time stuff means everything else has to take a step back to get deterministic behaviour. Having a laggy UI will totally get negative reviews even if the reason is “lab-grade” sound handling.
Hi, yes Filipo does a lot of RT-Kernel tweaks. Played around with different kernels when I built the Ian Canada streamer (Pi4 8GB RAM is perfectly fine).
I did have several conversations as I sponsor the Gentooplayer update server . Just ask him via his gmail address, he is always very responsive!
The kernel made some difference, of course it depends what services else you wanna activate and ones testet, put it into RAM system.
Looking not bad: