VIA Chipsets slow down PCI cards

Motherboards equipped with VIAs chipsets do not offer the full performance specified for PCI. VIA ist now offering a new patch to remedy this situation. In contrast to the first version, VIA seems to get things right this time.

We first reported about VIAs problems with PCI in early December 2001. When combined with a VIA-board, ATA/133 adaptors and fast SCSI cards can not live up to their performance expectations. In January 2002 VIAs first patch appeared on the firms support website, aimed at users of Promise cards. Of course we checked if this patch really helps and maybe solves problems with other cards.

VIAs problems with its PCI implementation surfaced with the move to Ultra-ATA/133. The maximum speed of the interface here increases to a theoretical maximum of 133 MBytes/s. Currently, Maxtor is also the only vendor to actually ship these drives, named DiamondMax D540X and D740X.

To fully use the speed of the interface of these drives, companies like Promise, Highpoint and ACARD offer PCI cards with Ultra-ATA/133-chips. These controllers fully consume the bandwidth of PCI: As with Ultra-ATA/133 PCI has a theoretical maximum of 133 MBytes/s. As with many other storage related units, vendors here define one megabyte to be one million bytes, and not 1048576 bytes. Therefore, in more technical units, the theoretical maximum is 127,2 MBytes/s.

In addition to that, Ultra160-cards for SCSI peripherals have long exceeded the limitations of the PCI bus, as their theoretical maximum is 160 MBytes/s.

In our tests with the maximum burst rate offered by todays hard drives, only motherboards equipped with chipsets by Intel, SiS and ALi showed decent performance. VIAs flaws were found with board designed for Intels and AMDs CPUs alike.