[NTLUG:Discuss] inconsistent sound

[Bryan J. Smith]

On Wed, 2004-10-27 at 04:08, Ralph Green, Jr wrote:
>   One system was an older VIA chipset motherboard(an MSI, like yours)
> running a Duron 950.  I tried several distros and they all had problems
> with the onboard sound.  I put an Audigy sound card in and it sounded
> fine.  Well, it sounded fine until a couple of years later when a
> lightning strike nearby killed the motherboard.

The Emu10k1 DSP drastically reduces the overhead on the PCI bus.

> On a Compaq Ipaq system with an Intel 810 chipset and a P3/650, I
> found some Linux distros had this problem and some did not(I did not
> keep notes on that aspect, so I can't tell you which had the problem).

The i810, like most chipsets, are high-integration, low-brains.  Even
though the audio is in the "southbridge," it's still connected to the
shared 0.125GBps PCI bus.  Sometimes it's not even a PCI master, but a
peer without interrupts, hence the major latency issues.

> I have seen a VNC server cause this problem, too.  Even though CPU
> utilization was not nearly maxed out,

Again, CPU utilization has *0* to do with interconnect.  Storage,
network and, in the case of desktops, audio are the *3* big items that
kill the measly 0.125GBps PCI bus.  Your 6.4GBps+ CPU Aggregate
Frontside Throughput (AFST) means _nothing_ to your measly 0.125GBps
which is offering 25x _less_ bandwidth.

Interconnect is everything, and it remained _unchanged_ since the i486
until PCI-Express came along (unless you went Opteron with an
AMD8131/8132 HyperTransport PCI-X tunnel, or Xeon with a ServerWorks
chipset for $300+ in just the mainboard).

[ BTW, AGP is a PCI bastardization (non-standard, Intel trade secret) on
its own, dedicated PCI bus.  Although AGP _can_ "contend" with the
"regular" PCI bus via the "northbridge" -- at least in Intel [A]GTL+
platforms (Socket-8 through 478).  PCI-Express also addresses this -- at
least on Athlon64/Opteron, although LGA-775 with DDR2 seems to be "good
enough" with 20 channels despite still being "bridged" AGTL+/FSB. ]

> stopping the VNC server on one machine is needed sometimes(XorbServer,
> or some such).  This was a Mandrake 10 system, and I don't know if
> the problem was anything specific to that version.

No.  VNC, like pcAnywhere, is remote framebuffer (RFB).  It sends
_every_ pixel individually.  This is the most _in_efficient remote
desktop access, and really hammers the video.

If your video is connected to the PCI bus, say goodbye to even
attempting any other I/O in the system.

> So, you may have lots of things to try.

Typically using a more advanced audio card, or 

In the case of newer systems, just getting the ATA and NIC off of the
shared 0.125GBps PCI (by using their own, dedicated PCI-Express x1
channels, even if on-"southbridge") now frees up _tons_ of bandwidth for
the audio.

Interconnect is everything.  10 years of only 0.125GBps for I/O (other
than video) -- that's sad.


-- 
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