VIA addresses PCI performance issues


— 12:00 AM on January 15, 2002

I had a conversation with VIA just now about the PCI performance problems recently pointed out by tecChannel, and I have a few answers for us. VIA is aware of the problem tecChannel unearthed, and they are working with manufacturers like Promise and Highpoint to help improve performance.

However—and they're mostly right about this—this issue affects performance only in very specific scenarios, like burst transfer rates off of fast drive arrays. So it's not the end of the world. (Heck, the KT266A and P4X266A are still two of the fastest chipsets in their respective classes.) Over the longer term, they say, we'll need 64-bit PCI or something similar in order to improve throughput, because the current mainstream PCI bus is simply getting overloaded.

VIA was also keen to point out that they hadn't had this problem brought to their attention before. Only now are multi-drive RAID arrays that can tax the PCI bus becoming common. Now that they're aware of the problem, VIA says they'll be working actively with PCI card manufacturers to help improve throughput. Apparently, these IDE RAID controllers address each PCI bus implementation a little bit differently, and there's room for improvement in how they handle VIA's PCI bus.

The first fruit of this effort is the mysterious patch for Promise RAID performance VIA released not long after tecChannel published. This Promise patch is not a copy of George Breese's PCI latency patch. It's intended specifically to improve PCI throughput with VIA chipsets and Promise RAID controllers.

As for George's patch, it attempts to reduce PCI latency with VIA chipsets. (Apparently, VIA's PCI bus implemention could benefit from latency reductions, as Ryu and others have suspected.) Unfortunately, though, some of the things it does are not fully compliant with the PCI 2.2 spec, so Microsoft wouldn't certify any VIA drivers that incorporated such tweaks.

That's the story so far. I'm a little too tired to sort through all the issues raised by these events right now, but I'm vaguely concerned that tecChannel found this problem before VIA's engineers did. Still, I'm pleased to know VIA is addressing the issue. We'll keep an eye on this one as it develops.

Tip: You can use the A/Z keys to walk threads.
View options

This discussion is now closed.