Results of Overclocking the PCI Bus
There has been much discussion regarding the benefits (or lack thereof) of overclocking the PCI bus, which potentially affects PCI and AGP cards, as well as IDE connected devices. The comparisons made are 33MHz vs. 37MHz PCI at a 75MHz bus speed, and 32MHz vs 41.5MHz PCI at 83MHz. The two motherboards used were the M Tech R581A which uses the SiS 5591 chipset and the AOpen AX5T rev 3.4 which uses the Intel 430TX chipset. The SiS 5591 chipset has very close to true asynchronous PCI at system bus speeds above 66MHz (at 83MHz the PCI bus runs at 32MHz rather than 33MHz). The R581A also implements another feature of the 5591 chipset, which allows switching between sync (37MHz) and async (33MHz) PCI operation at 75MHz via simple jumper change. Unfortunately, the 83MHz test was not quite as straight forward and required using a ‘baseline’ test of board boards at 66MHz to identify the differences between the chipsets and motherboard designs in regards to performance at the higher bus speeds.
|AMD K6-2||Winstone98 Business (Win95)||Winstone98 Business (WinNT)||High-End Winstone (WinNT)||Quake 1.06 (DOS 7)|
|75MHz Tests|| || || || |
|2.0×75 Async||15.7||15.7||15.4||11.1 fps|
|2.0×75 Sync||15.8||15.9||15.4||11.2 fps|
|4.0×75 Async||20.5||19.6||18.9||14.3 fps|
|4.0×75 Sync||20.4||19.4||19.1||15.4 fps|
|Baseline Tests|| || || || |
|2.5×66 (R581A)||16||16.1||15.6||11.3 fps|
|2.5×66 (AX5T)||16.3||16.3||15.9||11.1 fps|
|5.0×66 (R581A)||20.6||20.3||19.9||14.2 fps|
|5.0×66 (AX5T)||20.8||19.4||18.3||14.3 fps|
|83MHz Tests|| || || || |
|2.0×83 Async||16.9||16.9||16.5||12.0 fps|
|2.0×83 Sync||17||17||16.4||12.1 fps|
|4.0×83 Async||21.6||20.5||19.6||15.1 fps|
|4.0×83 Sync||22.0||20.9||19.8||17.0 fps|
These results clearly show that for the majority of applications, overclocking the PCI bus provides absolutely no benefit. This should have been fairly obvious, since the bandwidth of the PCI bus (cumulative) is 132MB/sec (at 33MHz) while IDE devices even at maximum burst speeds under UDMA can only achieve 33MB/sec transfer rates. In order for an increase in bus speed to actually have an effect, the devices attached would need to exceed the bandwidth available, which is clearly not possible under normal circumstances. However, you can also see that there is at least one situation where the definitely appears to be a benefit to overclocking the PCI bus.
The tests indicate that Quake frame rates increased by a noticable degree under two of the tests, but not under the others. It should be noted that both of these tests were at high CPU speeds. From these results we can surmise that 3D games such as Quake have a *huge* amount of I/O activity (both disk and video) which can actually saturate the PCI bus at the higher CPU speeds. A future set of tests will need to include an AGP video card to separate the disk I/O from the video I/O (AGP has twice the bandwidth of ‘standard’ PCI devices) to see if this removes the bottleneck at the lower PCI bus speeds.
Be the first to discuss this article!