PCI Delay Transaction/PCI 2.1 Compatible/PCI 2.1 Compliance
Options: Enable, Disable
PCI 2.1 specification mandates that this is *always* to be enabled. Strangely enough, it's disabled by default on far too many boards. This causes an ISA device--and you do still have some, even without ISA slots or cards, since the SuperIO chip does a lot of things on the ISA bus--to be told to wait if the PCI bus is in use. It's very much like Passive Release in the regard that it's a buffer between fast PCI and slow ISA.
ISA devices now use the PCI bus, with the PCI-ISA bridge translating the data. Since the ISA bus is so slow, telling them to wait is the best idea and it's not going to cause any delay to ISA transactions since PCI is so much faster.
HOWEVER, our good friend, the Sound Blaster Live, deserves special attention. The Live appears to not query the arbiter as to the status of "bus parking," and it is hypothesized that it incorrectly assumes a "last device" schema is in use, which is the default in most chipsets. For performance reasons, VIA always parks the bus on the arbiter, which results in faster switching between devices but higher latency for any single device. This option, if enabled, can cause SB Live cards to cause crashes on VIA chipset boards and performance penalties (including high sound latency) on Intel chipset boards. A VIA chipset always parks the bus on the arbiter, as previously mentioned, while an Intel chipset (440BX or better) will park it on the last device to have used it most of the time. Other cards, such as the Aureal Vortex2, can also be affected by this, but Aureal patched this up in later driver releases. It's only a real problem in the 32-bit environment of Windows 2000 and Windows XP with ACPI in use. When older methods of assigning IRQs are in use, the cards signal to the arbiter in a different way, bypassing this problem.
