There are many variables playing together. I would isolate things a bit, like testing it with one or two users or the single machines before going into network connections.
I think this thread is quite done since it is not an Ardour upgrading issue.
These tests showed that there are xruns even before upgrading to Ardour 6.5 using the Debian repos. (Though they are rare)
That means I can test further on my own. Since I know it worked better two years ago I could go back and test a vanilla Debian installation more intense. If there are issues it could be an issue caused by a changed hardware. If not it could be updates.
I try to post back here, when I find the cause for it.
Good luck for your setup. Maybe you open a separate thread for it.
I fixed the xrun issue and want to report as requested by @OConnorStP.
I fixed the issue by accident. Wanted to just try a realtime kernel because I stumbled across this.
I built a kernel from source, because I did not find an RT kernel with 1000Hz timer settings.
I took a config from Debian RT kernel package and patched the kernel source with realtime patches and configured it with 1000 Hz timer settings.
I was unable to test it because I had nVidia drivers running whivh donāt work with RT. I did not want to uninstall them and I was unable to get around this by adding kernel blacklisting parameters to grub config, so I just changed to my internal GPU that uses nouveau driver for the test of the RT 1000Hz kernel.
I tested that setup with Ardour 6.5 from Debian repos recording a loop section of two channels together with Guitarix running and MIDI raw driver active, MIDI bridge and Pulseaudio bridge running for at least six hours (six to eight typically) for four times. So I am a little more sure that this setup is stable before posting back here.
I donāt know if this is helpful for you.
I want to find out which of the things that I did really removes the xruns:
RT kernel? (Had 1000Hz setting already before so it is just the kernel, or config difference against liquorix)
GPU change? (Cannot really be, the GPU before worked perfectly
GPU driver change? (Might be, I found some reports that indicate they do not ibfluence system latency and some said they do)
The RT kernel could have been it, but like Paul said the 1000Hz setting seems to be irrelevant today, so using the RT kernel packaged by Debian should suffice if you find yourself doing a new install in the future. Did you also install the rtirq program to prioritize your soundcard? If not, that is recommended.
I had a desktop years ago with a NVIDIA card. It constantly spewed xruns until I installed the proprietary driver and then went into NVIDIAās settings panel and checked a box for āPerformanceā or something similar. After doing that, it behaved normally, so the GPU changes could have been it. I could only use a low-latency kernel with it, not a RT kernel, but it was adequate. I believe the Liquorix kernel plays nice with NVIDIA drivers, too, if you want to try it.
After that computer was retired, I have only bought computers with Intel HD graphics built-in because I donāt need a dedicated GPU, and it removes a layer of complexity to audio setup, assuming the situation today isnāt too different than it was several years ago.
A test with Liquorix and the onboard graphics card that uses nouveau driver shows xruns again (with the RT kernel there were no xruns). So nVidia drivers cannot be the problem here.
As soon as I find a way to quickly change to nouveau (blacklisting nVidia ideally, no uninstall) Iāll test if the RT kernel will work with the nVidia GPU and the nouveau driver to show that only changing the kernel to RT makes the difference, not the change of GPU or driver. But when I think about it - it is very unlikely that liquorix performs better with a dedicated nVidia GPU compared to the onboard GPU.
So I looks like currently in Debian I need an RT kernel which was not necessary before.
Lets find out right? I already tested the RT kernel from the Debian repositories two years ago they did not do the job. I report back how it performs.
No, I didnāt. Yeah maybe thats better, but it seems enough to have RT, 1kHz and WIFI off. There are so many other things that are suggested to make it better, but when there are no xruns with the current measures, why changing something?
I tried some of them some time ago, see some comments in Is it really possible to get down to 0 Xruns
Good to know!
I used the nVidia drivers together with liquorix and even tested it once for 6+ hours and I thought it works. Today it doesnāt anymore and I am not sure why. For now I just know that an RT kernel performs better.
I tried the official Debian RT kernel and it produces xruns. This kernel does not have the 1000Hz setting. It also not the best test, since I should have used my own build and adopt it but it consumes so much rime to build. Maybe later.
For building my RT kernel with 1000Hz I took the config of the stock RT kernel and changed the timers. So there should be only the timer settings that are different between stock and the kernel that has no xruns.
God bless you Robin. Iāve been dealing with shit DSP for months. I built a new computer much more powerful that my last and used the opportunity to upgrade to Mixbus7 and for months Iāve been trying to figure out with my DSP is worse than on my old comp. This one line of code fixed everything.