RME HDSP 9632 issues...or maybe not

3 replies [Last post]
iliverez
User offline. Last seen 2 years 25 weeks ago. Offline
Joined: 2010-10-12
Posts:

Hi,

I have an RME 9632 and I get randomly audible xruns, about a couple of them in a 4-6 minutes playback or recording session. Interesting enough, it makes almost no difference even if I compile the linux kernel at the same time. My typical setup is having a session with about 6 - 7 tracks, all muted apart from one and playing along with an edrum kit. Sometimes, I also record the kit.

Jack configuration is at 44100, 128 frames, 2 periods, or 44100, 256 frames, 2 periods.

I have tried almost anything, but I could not improve performance. Anything includes:

* AVLinux 4.2-5 (2.6.31-rt, 2.6.33-rt, 2.6.39, 3.0.x kernels)
* Fedora 16 with custom low latency 3.0.1 and 3.0.1-rt kernel)
* Tried various priority configurations following the tutorials found in ffado and linuxmusicians.com
* changing from jack2 to jack1 and vice versa
* using jack2 with or without the -S option

System specs are amd x2 5200+, 5gb ram, and a couple of TB hard disk space, nvidia 7900gs gpu. For the realtime kernels, I use the nouveau driver or the nv driver. I also use lightweight desktops (LXDE or enlightenment).

I guess 256 frames should be enough for the RME given the system specs and this kind of use. Given that in the future I will use more midi, I would even like to go down to 128 frames safely.

So if any 9632 owner can achieve better results, please provide some info for your configuration and system ;)

seablade
User offline. Last seen 2 hours 24 min ago. Offline
Joined: 2007-01-22
Posts:

I was receiving better results when I ran that card on Linux for years. I don't have a Linux box with it in it at the moment though to compare. I will say that most people I know sadly do get the best results with the binary driver, certainly instead of nouveau but in many cases even instead of nv IIRC. I know I always ran the binary driver and could get down to <3mS jack latency with no issue.

I would check things like making sure you aren't sharing IRQs etc. with anything else.

Seablade

iliverez
User offline. Last seen 2 years 25 weeks ago. Offline
Joined: 2010-10-12
Posts:

Thanks for the tip. I have already checked IRQs and even tried every available PCI slot on the motherboard. The results were just the same. I can't remember for sure, but I think at least at one slot I had a clean IRQ. I will check this again.

I have also tried setting the PCI bandwidth with setpci, without better results.

iliverez
User offline. Last seen 2 years 25 weeks ago. Offline
Joined: 2010-10-12
Posts:

For what is worth, after spending months compiling and tweaking kernels, settings etc. I 've figured out what caused the problems.

The xruns happen when I have the midi input of the 9632 physically connected with the midi output of the Roland TD-4 drum module. The relative midi port in jackd's universe is not connected with any other application. The audio stream comes from the drum module's audio outputs connected to the audio inputs of the 9632. This setup causes (at my system) xruns at any reasonable combinations of sample rate/frames for audio production.

Disconnecting the midi input of the 9632 or using another midi to usb cable for the midi connection of the drum module with the PC produced no xruns.

So with HDSP 9632 midi disconnected or using the usb to midi cable, I could go as far as 96kHz,128 frames,2 periods without xruns, tested with:
AVLinux 5, kernel 3.0.6 low-latency (32 bit, NVidia proprietary driver)
Fedora 15 with custom 3.0.1 rt kernel (64bit, nouveau driver)
Fedora 15 wiht custom 3.1 low latency kernel (64bit, nouveau driver)

If someone could run similar tests, e.g. with a keyboard that its audio output is connected to the soundcard's input and at at the same time the keyboard's midi out is connected at the same soundcard's midi input, please provide your test results.
A wild guess is that this hdsp's driver issue. A wilder guess is that this is an alsa/jack issue when both the audio and midi data come at the same time. Playing along a track of about 6 minutes should be enough to produce any xruns.