[Infowarrior] - Inte'ls new Sandy Bridge chips are buggy

[Richard Forno]

Intel finds flaw in 6-series chipsets, halts shipments
by Cyril Kowaliski and Scott Wasson — 10:07 AM on January 31, 2011
http://techreport.com/discussions.x/20326

As phenomenal as Intel's new Sandy Bridge processors turned out to be, nothing in this world is truly perfect. Intel announced earlier this morning that it has discovered a flaw in the 6-series chipsets that accompany the new processor family. While it reassures users that they can "continue to use their systems with confidence," the chipmaker has nonetheless halted chipset shipments until a new, bug-free version of the silicon starts to ship out late next month.

What's the problem? Intel explains, "In some cases, the Serial-ATA (SATA) ports within the chipsets may degrade over time, potentially impacting the performance or functionality of SATA-linked devices such as hard disk drives and DVD-drives."

For folks who have already crossed the Sandy Bridge, Intel adds that it will "work with its OEM partners to accept the return of the affected chipsets," and it plans to "support modifications or replacements needed on motherboards or systems."

Yes, that likely means the replacement of all Sandy-Bridge-based motherboards, laptops, and pre-built PCs currently on store shelves or already in the hands of consumers.

That sounds like a fair amount of hassle for all involved, but it probably beats the alternative—degraded storage performance on a state-of-the-art quad-core PC.

Beside the obvious inconvenience and bad PR, this little slip-up will cost Intel quite a bit of money, too. The firm expects to see a $300-million dent in first-quarter revenue (since full volume production of 6-series chipsets won't resume until April), not to mention $700 million in total repair and replacement costs.

Intel stockholders might not need to cut and run just yet, though. Intel claims it can make up for the lost revenue by year's end, and in the same press release, the chipmaker goes on to say it now expects first-quarter revenue to be in the $11.3-12.1 billion range, an increase from the previous forecast of $11.1-11.9 billion. Gross margin will, however, be understandably lower than initially expected (59-63% instead of 62-66%).

We are currently checking with Intel and motherboard makers to see how they plan to assist affected customers. Stay tuned for more info as we get it.

Update - 11:43 AM: Intel just held a conference call to talk about the Sandy Bridge chipset problems, and we now have a few more details to share with you.

The problem that's caused Intel to initiate a billion-dollar chipset recall affects the SATA ports on all 6-series chipsets, including the H67 and P67 chipsets most prominently used in consumer products. All of these chipsets are collectively referred to as "Cougar Point" inside of Intel. Because there are no third-party chipsets compatible with Sandy Bridge processors, all Sandy Bridge-based systems are potentially affected, including desktops, laptops, and BYOPC motherboards.

The issue is a circuit design problem resulting in a gradual degradation over time of SATA connectivity on the affected ports, manifesting itself as high bit-error rates on those ports and eventually as total device disconnects.

That's a serious issue, but it's limited in scope. Intel says storage devices connected to those ports should not be damaged, and data on the devices should be intact and readable on another system.

The ports potentially affected, interestingly enough, are the four 3Gbps SATA ports on the chipset. The two 6Gbps SATA ports aren't at risk.

Because this is a chip design-level problem, it will require the replacement of the Cougar Point chips embedded in the motherboards of affected systems.  Intel expects to be producing an updated, fixed version of Cougar Point silicon in late February, with "full volume recovery" coming later, in April or possibly even late March.  Implementing the fix will involve the replacement of a photomask for one of the layers of metal on the chip.  The layer in question is  apparently a "later" layer in the production process, so we expect there's some potential for partially completed chips currently in production to have the revised layer applied to them.  Note that the 6-series chipset is produced on Intel's very mature 65-nm fabrication process, not the cutting-edge 32-nm process on which Sandy Bridge CPUs are produced, so this isn't likely to be an especially thorny issue to untangle. Intel says the change should be "very straightforward" and it has "very high confidence" that the fix will be effective.

As you may know, Intel pours millions of dollars into validation testing for product like these, and its partners at major PC makers do the same. This problem apparently wasn't detected early on because of its nature, involving a slow degradation of SATA connectivity over time. Intel estimates that something like 5% of systems could develop problems over a three-year life span, assuming typical laptop usage patterns. Beyond that time window, the failure rate might rise further. For systems with heavier usage patterns, the failure rate during that initial three-year window could be as high as roughly 15%. That's obviously high enough to warrant the drastic action Intel is taking.

The first evidence of the problem cropped up during extended testing by PC makers, after the chipsets had passed the initial validation stages within Intel and within the OEMs. Intel says it learned of the problem last week; understanding and characterizing the problem then took a few days. That analysis concluded last night, and the company put shipments of its chipsets on hold this morning. From what we can gather, Intel partners were only very recently notified of the problem, too.

In addition to affecting systems already on the market, the chipset hiccup will delay the release of a host of laptops and other systems based on the dual-core variants of Sandy Bridge. Those systems were originally scheduled to begin hitting store shelves in the first couple of weeks of February, but Intel now estimates another "few weeks" will be added to those release schedules, depending on how long it takes PC manufacturers to incorporate the revised chipset silicon into their production pipelines. Intel's estimate sounds a little too optimistic to us, though. Given that the 6-series chipsets won't likely return to full production volumes until at least late March, we suspect the delays may add up to at least a couple of months in total.

This is obviously still a developing story, and we are working to understand how motherboard makers will address the problem for consumers.