[PlanetCCRMA] Latest ccrma FC3 kernel: 2.6.12-0.21.rdt.rhfc3.ccrma?

[Tracey Hytry]

Brad said, then Fernando said:

> I don't really know which one will have the best performance for audio
> work, probably the rdt kernel. 

I run FC3 and get good results on the machines with the 2.6.12-0.21.rdt.rhfc3.ccrma kernel.

I wouldn't mind trying out a newer one and put it through the paces on FC3 though.

We have been doing a lot of testing of kernels on FC4 both on this machine(athlon64 stock 32 bit FC4 and planet setup), and on a AMD dual core(abit AN8 ultra MB, 2 gig ram).  Besides testing the audio side of things(like the jack problems) we've been testing the compatibilty with the nvidia drivers.  We get most of the same results as Fernando has posted, and it can be very frustrating to see a high end machine go "to hell in a hand basket" trying to do low latency audio.

So far the 2.6.12-0.21.rdt for FC4 works about the same as the 2.6.13.. one that Fernando put out.  We have been getting a lot of problems with 2.6.14-0.10.rrt on FC4.  All of these have better audio latency then any stock FC4 kernel(2.6.12 through 2.6.14) if that helps?

You can pretty much forget about running the nvidia drivers on any kernel with premption because the nvidia driver seems to like to hold locks too often.  One of my more favorite lines is from the kernel where it tries to be polite with the nvidia driver:

kernel: agpgart: Putting AGP V3 device at 0000:02:00.0 into 8x mode
kernel: NVRM: Trying to sleep during raised irql!!
kernel: NVRM:   are we holding a lock?
kernel: NVRM:   skipping os_delay

Overall the 2.6.12-0.21.rdt on FC3 seems the be the most stable of everything so far.  I'm not sure about the new jack that was cooked up for FC4, as it doesn't seem to fix anything(and maybe makes more trouble).

I do have a question for Fernando:

When you compiled the FC3 and FC4 kernels did you do it from a machine running the distribution the kernel is made for?  Something in the back of my mind is giving me a suspicious feeling about FC4's gcc also, but that may be from seeing too many ghosts staying up too late testing kernels:)

Tracey.