Options : NA, 8, 1, 2, 3, 4, 5, 6, 7

The PCI bus is much faster than the ISA bus. So, for ISA cards to work properly with I/O cycles from the PCI bus, the I/O bus recovery mechanism adds additional bus clock cycles between each consecutive PCI-originated I/O cycles to the ISA bus.

By default, the bus recovery mechanism adds a minimum of 3.5 clock cycles between each consecutive 8-bit I/O cycle to the ISA bus. The options above enable you to add even more clock cycles between each consecutive 8-bit I/O cycle to the ISA bus. Choosing NA sets the number of delay cycles at the minimum 3.5 clock cycles.

So, set the 8-bit I/O Recovery Time at NA if possible for optimal ISA bus performance. Increase the I/O Recovery Time only if you are having problems with your 8-bit ISA cards. Note that this function has no meaning if you are not using any ISA cards.

16-bit I/O Recovery Time

Options : NA, 4, 1, 2, 3

The PCI bus is much faster than the ISA bus. So, for ISA cards to work properly with I/O cycles from the PCI bus, the I/O bus recovery mechanism adds additional bus clock cycles between each consecutive PCI-originated I/O cycles to the ISA bus.

By default, the bus recovery mechanism adds a minimum of 3.5 clock cycles between each consecutive 16-bit I/O cycle to the ISA bus. The options above enable you to add even more clock cycles between each consecutive 16-bit I/O cycle to the ISA bus. Choosing NA sets the number of delay cycles at the minimum 3.5 clock cycles.

So, set the 16-bit I/O Recovery Time at NA if possible for optimal ISA bus performance. Increase the I/O Recovery Time only if you are having problems with your 16-bit ISA cards. Note that this function has no meaning if you are not using any ISA cards.

Passive Release

Options :  Enabled, Disabled

Because the ISA bus is so much slower than the PCI bus, any device engaging the ISA bus normally renders the PCI bus inaccessible to the CPU. But with Passive Release enabled, CPU-to-PCI bus accesses are allowed via passive release of the PCI bus. It does so with the use of the chipset's embedded 32-bit posted write buffer.

The write buffer stores the PCI writes from the CPU when the ISA bus is being accessed. The data is then written to the PCI bus by passive release. Therefore, the processor can still access the PCI bus even when the ISA bus is being accessed.

Otherwise, the arbiter will only accept another PCI master access to local DRAM. In other words, only another PCI bus master can access the PCI bus, not the processor. This feature is used to meet the latency of the ISA bus master, which is much longer than that of the PCI bus master.

Enable Passive Release for optimal performance. Disable it only if you are facing problems with your ISA cards.

Delayed Transaction

Options : Enabled, Disabled

This feature is used to meet the latency of PCI cycles to and from the ISA bus. The ISA bus is much, much slower than the PCI bus. Thus, PCI cycles to and from the ISA bus take a longer time to complete and this slows the PCI bus down.

However, enabling Delayed Transaction enables the chipset's embedded 32-bit posted write buffer to support delayed transaction cycles. This means that transactions to and from the ISA bus are buffered and the PCI bus can be freed to perform other transactions while the ISA transaction is underway.

This option should be enabled for better performance and to meet PCI 2.1 specifications. Disable it only if your PCI cards cannot work properly or if you are using an ISA card that is not PCI 2.1 compliant.

PCI 2.1 Compliance

Options : Enabled, Disabled

This is the same thing as Delayed Transaction above.

This feature is used to meet the latency of PCI cycles to and from the ISA bus. The ISA bus is much, much slower than the PCI bus. Thus, PCI cycles to and from the ISA bus take a longer time to complete and this slows the PCI bus down.

However, enabling Delayed Transaction enables the chipset's embedded 32-bit posted write buffer to support delayed transaction cycles. This means that transactions to and from the ISA bus are buffered and the PCI bus can be freed to perform other transactions while the ISA transaction is underway.

This option should be enabled for better performance and to meet PCI 2.1 specifications. Disable it only if your PCI cards cannot work properly or if you are using an ISA card that is not PCI 2.1 compliant.

AGP Aperture Size (MB)

Options : 4, 8, 16, 32, 64, 128, 256

This option selects the size of the AGP aperture. The aperture is a portion of the PCI memory address range dedicated as graphics memory address space. Host cycles that hit the aperture range are forwarded to the AGP without need for translation. This size also determines the maximum amount of system RAM that can be allocated to the graphics card for texture storage.

AGP Aperture size is set by the formula : maximum usable AGP memory size x 2 plus 12MB. That means that usable AGP memory size is less than half of the AGP aperture size. That's because the system needs AGP memory (uncached) plus an equal amount of write combined memory area and an additional 12MB for virtual addressing. This is address space, not physical memory used. The physical memory is allocated and released as needed only when Direct3D makes a "create non-local surface" call.

Win95 (with VGARTD.VXD) and Win98 use a "waterfall effect". Surfaces are created first in local memory. When that memory is full, surface creation spills over into AGP memory and then system memory. So, memory usage is automatically optimized for each application. AGP and system memory are not used unless absolutely necessary.

Many people recommend the AGP aperture size should be half of the amount of RAM you have. However, that's wrong for the same reason why swapfile size shouldn't be 1/4 of the amount of RAM you have in your system. As with the swapfile's size, the AGP aperture size required will be smaller as the graphics card's memory increases in size. That's because most of the textures will be stored on the graphics card itself. So, graphics cards with 32MB of RAM or more will require a smaller AGP aperture than graphics cards with less RAM.

If your graphics card has very little graphics memory, then you should set as large an AGP aperture as you can, up to half the system RAM. For cards with more graphics memory, you shouldn't set the aperture size to half the system RAM. Note that the size of the aperture does not correspond to performance so increasing it to gargantuan proportions will not improve performance.

Still, it's recommended that you keep the AGP aperture around 64MB to 128MB in size. Now, why is such a large aperture size recommended despite the fact that most graphics cards now come with large amounts of RAM? Shouldn't we just set it to the absolute minimum to save system RAM?

Well, many graphics card require at least a 16MB AGP aperture size to work properly. This is probably because the virtual addressing space is already 12MB in size! In addition, many software require minimum AGP aperture size requirements which are mostly unspecified. Some games even use so much textures that AGP memory is needed even with graphics cards with quite a lot of graphics memory (32MB).

And if you remember the formula above, the amount of AGP memory needed is more than double that of the required texture storage space. So, if 15MB of extra texture storage space is needed, then 42MB of system RAM is actually used. Therefore, it makes sense to set a large AGP aperture size in order to cater for every software requirement.

Note that reducing the AGP aperture size won't save you any RAM. Again, what setting the AGP aperture size does is limit the amount of RAM the AGP bus can appropriate when it needs to. It is not used unless absolutely necessary. So, setting a 64MB AGP aperture doesn't mean 64MB of your RAM will be used up as AGP memory. It will only limit the maximum amount that can be used by the AGP bus to 64MB (actual usable AGP memory size is only 26MB).

Now, while increasing the AGP aperture size beyond 128MB wouldn't really hurt performance, it would still be best to keep the aperture size to about 64MB-128MB so that the GART table won't become too large. As the amount of onboard RAM increases and texture compression becomes commonplace, there's less of a need for the AGP aperture size to increase beyond 64MB. So, it's recommended that you set the AGP Aperture Size as 64MB or at most, 128MB.