Once Pipewire was available, I jumped at that. It wasn't even stable at the time, but I never want to fiddle around with Jack settings again. Now it all just works effortlessly in the background. I love how much I don't have to think about it.
@@foss_sound "..fully implemented in the pro apps". This will likely not happen. PW wasn't even originally supposed to have it's own layer, but for whatever reason that was still created. This is what Paul Davis (Product Manager of Ardour) told me, and even gave me explicit permission to quote him: "you can quote on me no native PW for ardour, certainly" :)
@@SudoMetalStudio yeah, paul is pretty strict with his view and if he is doing things. Sometimes I wish Justin Frankel (Reaper) inspired hin even more ... He said "reaper natively on linux will never happen", few years later he is working with linux, recording with linux and reaper is available for linux. And he copied jacks routing for reaper. :D (and loves jack)
About the LL kernel: I made round trip tests with my Focusrite 2i2. I got no difference to the vanilla kernel (Debian 12 stable and Fedora WS 38 at this time). So I adjust the config for audio user group (rtprio, memlock) and I never had any issues and good latency. Not as good with RME ASIO drivers at Windows 10, but 64 Samples is good enough for my recordings ... mixing is way lower sample buffer ... mostly use 512 and more.
I also read somewhere that most of the LL kernel optimizations can already be done with vanilla kernel params, so I think LL variant won't even be necessary in the future. But what I've seen and heard others doing, many say they got significant improvement to latency when switching to LL variant. And yea, I don't think any generic optimizations can ever beat manufacturers own drivers. But like you see in the end of the video, I get 6,7ms roundtrip latency with 64@48kHz which equals to "playing 2,3m further from amp". If you can half the latency (which is not even realistic), you'd only stand 1m closed to the amp. That's a difference no human being can distinguish :)
@@foss_sound I have RTX 3060 and I haven't had any problems with it. Except that NVidia is not supported by the Wayland version of KDE Plasma that comes with Ubuntu Studio 24.04, so have to stick with X11 version. I know the NVidia hate in Linux community due to the proprietary bullshit, but for me it has been a solid experience. I even run local Stable Diffusion and a local GPT without any issues, obviously heavily utilizing GPU.
@@SudoMetalStudio I never had problems with nvidia, except with the LL kernel. Sure thing I'd prefer foss drivers, but I am not that dogmatic. ATM I only use my thinkpad and just have no desktop pc with seperate GPU. But when I'll get one time soon, it will probably be with AMD CPU and GPU.
Hyvin epämystifioitu! Noticed the same thing when testing jack vs. pipewire-jack with LSP latency meter. ~0.2ms in PWs favor while average RTT was around 10ms. Unperceivable in use but props to PW team for a very solid implementation.
@@ospifi jack2 gets pretty much on par when used in async mode (sync is default I believe). In sync mode the buffer is tripled (1 x input + 2 x output) to ensure click-free connections, but in async mode the output is not doubled. This means the DSP latency is 33% better in async mode with jack2. Because PW shows similar results, I assume it's just doing similar single buffering in both input and output.
Greetings from Brazil, dude! Just so you know where your good work is reaching. Thank you for this video!
Perfect presentation, and a few things I was not aware of have now been added to my internal head library :)
Once Pipewire was available, I jumped at that. It wasn't even stable at the time, but I never want to fiddle around with Jack settings again. Now it all just works effortlessly in the background. I love how much I don't have to think about it.
Yeah, PW will give Linux pro audio a bump. Someday, when it's "done" and fully implemented in the pro apps.
@@foss_sound "..fully implemented in the pro apps". This will likely not happen. PW wasn't even originally supposed to have it's own layer, but for whatever reason that was still created. This is what Paul Davis (Product Manager of Ardour) told me, and even gave me explicit permission to quote him: "you can quote on me no native PW for ardour, certainly" :)
@@SudoMetalStudio yeah, paul is pretty strict with his view and if he is doing things. Sometimes I wish Justin Frankel (Reaper) inspired hin even more ... He said "reaper natively on linux will never happen", few years later he is working with linux, recording with linux and reaper is available for linux. And he copied jacks routing for reaper. :D (and loves jack)
About the LL kernel: I made round trip tests with my Focusrite 2i2. I got no difference to the vanilla kernel (Debian 12 stable and Fedora WS 38 at this time). So I adjust the config for audio user group (rtprio, memlock) and I never had any issues and good latency. Not as good with RME ASIO drivers at Windows 10, but 64 Samples is good enough for my recordings ... mixing is way lower sample buffer ... mostly use 512 and more.
I also read somewhere that most of the LL kernel optimizations can already be done with vanilla kernel params, so I think LL variant won't even be necessary in the future. But what I've seen and heard others doing, many say they got significant improvement to latency when switching to LL variant.
And yea, I don't think any generic optimizations can ever beat manufacturers own drivers. But like you see in the end of the video, I get 6,7ms roundtrip latency with 64@48kHz which equals to "playing 2,3m further from amp". If you can half the latency (which is not even realistic), you'd only stand 1m closed to the amp. That's a difference no human being can distinguish :)
@@SudoMetalStudio The LL kernel wrecked my nvidia drivers back then, ever since, i am fine without.
@@foss_sound I have RTX 3060 and I haven't had any problems with it. Except that NVidia is not supported by the Wayland version of KDE Plasma that comes with Ubuntu Studio 24.04, so have to stick with X11 version. I know the NVidia hate in Linux community due to the proprietary bullshit, but for me it has been a solid experience. I even run local Stable Diffusion and a local GPT without any issues, obviously heavily utilizing GPU.
@@SudoMetalStudio I never had problems with nvidia, except with the LL kernel. Sure thing I'd prefer foss drivers, but I am not that dogmatic. ATM I only use my thinkpad and just have no desktop pc with seperate GPU. But when I'll get one time soon, it will probably be with AMD CPU and GPU.
Hyvin epämystifioitu! Noticed the same thing when testing jack vs. pipewire-jack with LSP latency meter. ~0.2ms in PWs favor while average RTT was around 10ms. Unperceivable in use but props to PW team for a very solid implementation.
@@ospifi jack2 gets pretty much on par when used in async mode (sync is default I believe). In sync mode the buffer is tripled (1 x input + 2 x output) to ensure click-free connections, but in async mode the output is not doubled. This means the DSP latency is 33% better in async mode with jack2. Because PW shows similar results, I assume it's just doing similar single buffering in both input and output.
Really inspiring and professional as well. Well done!
Welcome back it has been a while! :D
@@linusbergsman2832 Yea, summer was very busy. But already working for the next video. It's going to be a BLAST when I'm done 🎶💪😉
@@SudoMetalStudio Nice dude hopefully we'll grab a beer someday together again! 😁