VIA Chipsets slow down PCI cards

ATA/133: IDE-Timing

To check the timing within the IDE bus itself we attached the cards by Promise and Highpoint to our Tektronix TLA704 Logic-Analyzer . Despite the bad performance with burst transfers we found the actual cycle timing to be correct. The Promise Ultra133 TX2 and Highpoints RocketRAID 133 transfer data within cycles of 30ns as per the specification of ATA/133.

If this is so, why do VIA motherboards only give you 64 to 90 MBytes/s effectively? The answer lies within the many breaks these systems take during bursts.

Normally, a burst should be performed continuously and without any interruption. The above pictures shows a burst like this, recorded using the Promise Ultra133 TX2 and and Intel-845-motherboard. The whole burst here is performed correctly. The length of a burst, however, can differ depending on the data transferred.

With VIA boards, this high speed transfer from the cache of a hard drive is constantly interrupted within a couple of µs. It has then to be re-initiated, as the next picture shows. Therefore, the effective burst rate drops to 64 to 90 MBytes/s at best.

In contrast to chipsets from Intel, SiS and Ali, VIAs products seem to have difficulties with maintaining high transfer rates close to the maximum speed of PCI for a longer time. As with ATA/133 PCI has a theoretical maximum bandwidth of 127,2 MBytes/s. It seems fair to conclude that VIAs implementation of a PCI bus must be the reason for the problems found.

To check the signaling with a VIA chipset we used VIAs P4XB motherboard. It is equipped with VIAs P4X266A chipset, using VT8233 as a southbridge. This VIA chipset was the best of the worst when we measured the transfer rates with ATA/133 bursts.