[NTLUG:Discuss] inconsistent sound

[Bryan J. Smith]

On Tue, 2004-10-26 at 22:20, Kevin Brannen wrote:
> If I didn't know better, I'd say it was bogged down by some other 
> process like updatedb, but it ain't so.  I'll run it with xosview up so 
> I can see instantaneous changes, and the CPU is barely being used, and 
> the disk is only being used in minimal amounts.  It's a single AMD 
> 2.4Ghz Athlon with an AC97 chip on an MSI MB w/ 1G of RAM, running Suse 
> 9.0; so I'm having trouble imagining this isn't a powerful enough 
> system.

Power isn't I/O.  If you've got something taxing the PCI bus -- which is
only a measly 0.125GBps (32-bit @ 33MHz) _ideal_, then that could be an
issue.  I've had the same issue under Windows and Linux, if something
bursts through the PCI bus, the sound may get squeaky.

It could also be a sound server issue.

> Especially since my old 500Mhz P3 never had this problem,

Ironically, nearly all Socket-478 and 462 mainboards don't have any
better PCI -- maybe 0.25GBps (64-bit slots) if you're lucky.

Fortunately, PCI-Express solves the problem nicely without having to go
the more costly PCI-X route, although a few Opteron mainboards will
offer both.

> of course it had a SBLive! in it.  I'm wondering if the builtin AC97 is the 
> problem, but that seems strange.

It doesn't at all.  It all depends on how much host processing you put
on your system CPU and memory, which then hogs up the I/O.  Cheap audio
puts more on the CPU, where the audio chip itself is little more than a
DAC.

Same issue with software RAID-5, it's not the XORs, it's the taxing of
the I/O bus in sending in all data to the CPU and memory for the XORs to
be calculated.  That's why an XOR engine on-card is better, because it
doesn't send the data over the PCI bus.

Over a measly 0.125GBps PCI interconnect, it doesn't matter if you have
1.6-3.2GBps to DDR memory, possibly (2) _true_ DDR channels (Socket-604,
939/940), or a 6.4-8.0GBps "front side bottleneck."  The I/O getting
roasted at 1/10-1/50th the CPU-memory.

> It doesn't waver all the time, but it's almost guaranteed to do it at
> least once per 10 songs, sometimes more.
> Has anyone heard of this before, or does anyone have thoughts on what to 
> look for?

Yes, I have heard it in games in Windows as well as when doing a lot of
disk I/O in either, and all you have is a single 0.125GBps PCI channel.

It's the #1 reason why RISC/UNIX systems still kicked the pants off of
most PC Servers, at least until the Itanium and, far more cost
effective, Opteron came along.

Luckily for consumers, PCI-Express solves it for a lot cheaper than
PCI-X.  Mainboards are already out for the LGA-775, and ones for
Socket-939/940 hit next month.


-- 
Bryan J. Smith                                  b.j.smith at ieee.org 
------------------------------------------------------------------ 
"Communities don't have rights. Only individuals in the community
 have rights. ... That idea of community rights is firmly rooted
 in the 'Communist Manifesto.'" -- Michael Badnarik