A Bug Bites Intel's Lindenhurst

This time it involves "Lindenhurst," the core-logic chipsets that work with Intel's new 64-bit Xeon processor (code-named Nancona). (There are two Lindenhurst chipsets -- the parts E7320 and E7520. The former supports fewer I/O ports and is cheaper.)

Prior to the release of Nancona at the end of June, the Xeon product line was all 32-bit. As a hybrid 32/64 bit CPU, which can run both 32- and 64-bit software via a set of 64-bit instruction-set extensions, Nancona is a significant step forward for Intel. Equally important, it marks Intel's response to AMD, which created the 32/64-bit category when it released its Opteron server CPU in April 2003.

The Lindenhurst chipset takes standalone Nancona processors and turns them into working servers by connecting the CPU to the system bus, memory and I/O.

At the time Nancona was unveiled, Intel simultaneously released a chipset (the E7525) to enable systems builders to put together workstations using the processor.

Earlier this week, Intel announced the Lindenhurst chipsets, which enable the construction of single- and dual-processor Xeon servers.

But there's one catch. As VARBusiness senior editor Jeffrey Schwartz has reported, there's a bug in Lindenhurst. Namely, in certain situations, the chipsets can't recognize PCI Express cards plugged into the server.

That's not a good thing for Intel, which is touting PCI Express as one of the biggest benefits of the Nancona platform. PCI Express is Intel's latest interconnect technology -- what we used to call a bus in the old days, except technically PCI Express isn't a bus. It's a switched, point-to-point connection. PCI Express is billed as delivering 3.5 times more bandwidth than the current, plain-vanilla PCI bus.

The Lindenhurst glitch didn't stop Dell, HP and IBM from announcing Nancona-based servers earlier this week. On the other hand, as Jeff reported, Gateway has decided to wait until Intel fixes the bug with new Lindenhurst silicon scheduled for release in the fourth quarter of this year.

Intel itself uses very careful language in talking about the problem. "Right now, we're not recommending that vendors use the external PCI Express slots," an Intel spokesman told me.

He said that Intel found the glitch in lab testing of Lindenhurst, but that "it never actually happened in all of our OEM tests."

Asked why Intel released a part if there's a problem with it, he said: "Because there's still a great deal of value that can be gotten from the chipset. This errata happens only in rare instances."

The instances are indeed arcane. The technical specifics can be found in Intel's specification update for Lindenhurst. (Spec updates are essentially bug lists, which Intel began releasing regularly in the wake of its Pentium FDIV flaw of 1994.)

Poring through the spec updates, I found three or four that appeared related to the PCI Express problems. One indicated that time-outs sometimes occur, another noted something called "compliance mode issues." Most interesting was a glitch entitled "PCI Express add-in card presence detect state misreported." This one described a problem opposite to the one discussed publicly. It said that "software may interpret the presence of an add-in card where none exists."

Equally surprising was something I uncovered that, in another context, might be considered old intelligence. Namely, the E7525 workstation chipset released for Nancona in June, which hasn't been the subject of press reports, exhibits exactly the same glitches that plague Lindenhurst.

As we drill down, perhaps deeper than we want to, are there any lessons here? For systems builders and VARs, the main takeaway is that you don't want to be first out of the gate with any new technology. Leave it to Tier 1's like Dell, HP and IBM to buzz out glitches, such as those surrounding the new Xeon chipsets.

The resolution of such growing pains typically takes only a couple of months, so any time-to-market issues you're worried about will be more than made up for by the comfort you can take in knowing that you're probably avoiding some killer customer support issues.

As far as Intel is concerned, it seems possible that part of its impetus for announcing the Xeon chipsets now may have been to blunt the advance of AMD's Opteron processor. Intel presumably did a cost/benefit analysis and decided it was better off getting an initial product out to market and fixing any minor errata later. The alternative -- waiting -- might have been seen as giving AMD time to grow stronger.

A few months' hence, when Intel fixes today's problems with revised Lindenhurst chipsets, or when the next inevitable minor glitch crops up in some unrelated part, this whole to-do will fade from memory.

So is L'Affaire Lindenhurst a cautionary lesson about the manner in which marketing considerations sometimes seem to trump technology, or is it much ado about nothing?

Perhaps, paradoxically, the answer to both those questions is yes.