Re: [linux-audio-user] AMD64 question

On Tue, 2006-06-27 at 06:17 -0700, Mark Knecht wrote:
> Hi Dave
>
> On 6/27/06, Dave Phillips <dlphillips@email-addr-hidden> wrote:
> > Greetings:
> >
> > Not long ago I mentioned that a student had traded an MSI mobo (socket
> > 939) for some lessons. I'm ready to start building a system around that
> > board, and I have some questions for this list:
> >
> > 1) I can get a new AMD64 Athlon 3800 2.4 GHz for (US) $145. Is the
> > Athlon 64 a good chip for audio work, and is that a good price ? It's
> > the best listed on Pricewatch.
>
> Yes. Im using an AMD64 Athlon 3000+. Works fine. You might want to
> check in with Lee Revel or someone technically in tune with the kernel
> developers about dual core chips. Mine is single core. There are still
> (I think...) soome dual core issues but they are fewer now then
> before.

I finally have an AMD64 box so I can speak from experience ;-)

I think the broken timekeeping on dual core machines has been sorted out
in 2.6.17. As gettimeofday() has to use the ACPI PM timer due to buggy
TSCs, it is slower and less precise than it could be, but at least the
time does not jump backwards and forwards anymore. I'm hoping AMD will
fix this bug (and AMD has acknowledged that it is a bug) in future
generations of chips.

One side effect that is still unresolved is that the latency tracer does
not work as it uses rdtsc directly - the reported latencies creep
inexorably upward as the tracer switches between cores. I am working on
a solution, but it's a real pain.

I've been running rt-exec on this machine
(ftp://ftp.compro.net/public/rt-exec) to test it and the max reported
latency is about 400usecs. That's pretty high for an -rt kernel but
good enough for audio work. I can't tell which code path is responsible
until I get the latency tracer working.

Lee
Received on Tue Jun 27 20:15:25 2006