I developed a program that allows adjustment of the pci latency setting of
the agp video card for the benefit of those using their pc for making music.
Reducing the agp latency setting from 248 to 128 often is a big help for the
pci sound card.
The problem I have is I don’t have a specific idea of why this should be. I
note that the agp bus is electrically distinct from the PCI bus that
contains the sound card, and the agp has its own direct path into memory and
so forth. Agp does, of course, expose some of the PCI configuration
settings such as that of the latency timer at 0x0d into the configuration
header.
So I can speculate that when the agp bus is going balls-out transferring
data using its high-bandwidth technology, that devices on other PCI devices
are somewhow prevented from acquiring some shared resource. Thing is, I
wish I had more specific knowledge of the exact nature of this performance
contention.
If anybody can help, it is greatly appreciated. I have the Shanley book on
PCI, but it explains the workings of individual devices on a shared bus and
I can’t find a reason why two busses would have contention–memory access
would be the number one suspect.