Re: The 850 chipset had a problem with PCI graphics cards, not all non-bus-mastering PCI cards.
P4 chipset bug affects more than graphics
--------------------------------------------------------------------------------
posted 1:53pm EST Thu Dec 28 2000
NEWS
Yesterday, we posted a news item about a bug in the i850 chipset, the only chipset that is currently available to support the Intel Pentium 4 processor. Intel delayed the Pentium 4 from October 30th to November 20th to examine a bug in the i850 chipset. What happened after that is unclear. Initial reports in September suggested that the bug only happened to specific PCI graphics cards when used with an AGP graphics card. Now it appears that either the bug has more effects, or it has changed its behavior due to a change in the i850 chipset.
Techweb reports that, "Specifically, passing large amounts of data over the PCI bus can introduce latency -- or computing delays -- into the system." The CRN (Computer Reseller News, another CMP publication) Test Center ran into the problems while benchmarking a DVD decoder card in shipping Pentium 4 systems. The AC-3 sound data was badly broken up, but, interestingly, older Pentium 4 test systems did not display the defect.
The article also states, "When asked if streaming other high-bandwidth data types on the PCI bus, such as between SCSI cards, might also inspire system latency, Intel indicated that bus mastering devices would not be subject to the problem." Thus, the current understanding is that bus mastering PCI cards will not cause latency, but non-bus mastering PCI cards might.
Read more at TechWeb.
ROB'S OPINION
Well, this just seems to keep getting worse the more I read about it. First, it only affects dual video with an AGP/PCI card setup. Now, it's affecting PCI DVD decoder cards. Will PCI sound cards be affected as well?
Back in September--if you reread the reports, as I did--you'll find that it was reported that Intel fixed the bug in the chipsets it was producing then. It appears now that the "fix" either was ineffective or caused more problems. The CRN reports echo this with the DVD decoder issue not showing up in the first P4 beta systems it tried, but showing up in the shipping P4 systems it tested on. It looks like Intel hasn't learned from its Pentium III 1.133GHz recall. Perhaps it was completely oblivious to the bug until now.
So far, the good news for P4 owners is that the problem only affects non-bus mastering PCI cards. I recall that the PCI network and SCSI cards I've worked with are bus mastering, but there's always the chance that off-brand vendors would skip the bus mastering. It appears from the reports that graphics cards and DVD decoder cards are not bus mastering. Bus mastering support is one of those things that you are not supposed to need to know about. PCI was set up to be Plug-and-Play, but back in the early days some PCI slots didn't support bus mastering and you had to monkey around with your PCI cards and put them in different slots. Nowadays, most motherboards support all bus mastering slots, but some cards may not be bus mastering, and those cards, it appears, may wreak havok with Pentium 4 systems with the shipping i850 chipset. We'll have more details as they arrive.
I just checked out a couple of vendor specifications for PCI cards,and didn't find any reference to whether they were bus mastering or not. That's going to make this issue tough to deal with. If you want to read up on bus-mastering check out PCGuide.
geek.com |
