Intel shakes off $1bn chipset flaw

How much did a little design flaw in one metal layer on the Cougar Point Intel 6 chipset used with Sandy Bridge Core i5 and i7 chips cost Intel? A cool $1bn, that's how much.

And in a conference call with Wall Street analysts on Monday, Intel's two Smiths – Stacey Smith, Intel's chief financial officer, and Steve Smith, director of PC/client operations and enabling - shook off the hit like it was no big deal.

As El Reg previously reported, Intel announced first thing Monday morning that the Cougar Point chipset used with the quad-core Core i5 and i7 chips had a flaw that could cause their Serial-ATA (SATA) ports to degrade over time – and even fail. The company is taking a $300m hit on its books to cover the manufacturing costs of the faulty parts, and will book another estimated $700m in charges relating to replacing the motherboards that have been equipped with the faulty chipsets.

CFO Smith said that Intel had shipped slightly fewer than eight million of the Cougar Point chipsets, which are used to link the I/O peripherals on a system motherboard to the CPU. Most of these have been already welded onto motherboards, and that $700m is what Intel thinks it will cost the company to pay for the replacement of those motherboards among its OEM partners, who are no doubt not going to pay for remaking the mobos because this is Intel's fault.

VP of PC chips Smith added that unit shipments to customers of the faulty parts were small – just over 100,000 parts – and that the faulty circuit design did not cause immediate failure or even the risk of imminent failure.

"To the best of my knowledge, there has not been a single consumer report with this issue," Smith the VP of PC chips said on the call.

The exact nature of the flaw was not disclosed, but Smith the chip guy said that the Cougar Point chipset had made it through Intel's "rigorous" testing regimen, and had similarly made it through the QA processes of its OEM partners.

But after the January launch of the Sandy Bridge Core chips, an OEM did some more intense testing and found high bit-error rates on the legacy 3Gb/sec SATA ports implemented by the chipset. The chipset has four of these 3Gb/sec ports as well as two 6Gb/sec SATA ports, the latter of which are not affected by the flaw.

If you happen to have a shiny new Sandy Bridge laptop, you might see high error rates on these SATA ports, and then one day you might see the device simply disappear. However, Intel says that the flaw will not corrupt data – it will merely mean the laptop won't see the SATA device once it fails.

Intel's own QA people confirmed the Cougar Point design flaw and started testing the chipset early last week, and took a few days to characterize what this might mean for end users. As it turns out, the effect is that on a normally used laptop or one that runs in a normal environment, there is about a 5 per cent chance of failure in the typical three-year span. A heavily used machine or one in a hot environment could see as much as a factor of three higher chance of failure, according to Smith (still the chip guy). Lightly used Sandy Bridge laptops would see a lower failure rate, based on Intel's models. This final analysis of failure rates was done last night, and Intel made the announcement this morning after Wall Street opened for business.

"It's been a very rapid process," said Smith. The chip Smith.

And it is just a coincidence that Intel closed its deal to acquire the wireless chip business of Infineon this morning and said it was pretty sure that the McAfee deal would close before the end of the first quarter, giving it a chance to add in the revenues from these two companies, which have a combined annual run rate of about $3bn, according to Smith, the numbers guy.

As it turns out, Intel's forecasts for the first quarter include about two months of revenue from Infineon and one month from McAfee, which is why revenues are projected to rise by $200m to $11.7bn (plus or minus $400m), up from the prior guidance of $11.5bn (with the same plus or minus). Intel is taking the $300m revenue hit to replace the Cougar Point chips, but is making it up with $500m in acquisitions and adjusting profits downward in the fourth quarter.

Nothing to see here, move right along.

The Cougar Point chipset is manufactured in Intel's old 65 nanometer process, which it can ramp very fast, and the flaw in the chipset is only in the upper metal layers of the chip, so masks for lower layers are unchanged and a large number of chips already in production can be tweaked with the design change Intel has put in its masks.

Smith the numbers guy reiterated that the redesigned Cougar Point chipset would start shipping to OEMs at the end of February, and that the chipset would ramp to full volume by April.

This is what Intel said in its statement early Monday morning. However, Intel's CFO added that there is a good chance that volume production for the chipset could occur in March, and OEMs could get the chips in time and put them into new motherboards quickly enough to not put much of a dent in OEM's Sandy Bridge laptop rollout.

Numbers Smith also said that Intel had inventory of older Core i5 and i7 chips and their chipsets to fill in the gap, and that there would probably be a couple of million units in the first quarter where demand won't be met. However, he was convinced that Intel would be able to make up the volume early in the second quarter and that the net effect would be nil.

"Our volume will be robust in Q2 and beyond," Smith-numbers said, adding that the Sandy Bridge chips will still be baked and sold along their normal ramp, and that Intel expected these quad-core products as well as the upcoming dual-core products to be the keystones to Intel's and its OEM partners' "back-to-school" sales later this year.

Those mainstream, two-core Sandy Bridge parts were due to be announced in two weeks. Intel is now working with the laptop OEMs to figure out how to match up the deliveries of the new Intel 6 chipsets with motherboard and system manufacturing to reschedule that launch. "I think we will recover pretty quickly," the CFO said.

The design flaw has been fixed in chipsets that are derivatives of Cougar Point that are used for servers, desktops, and workstations.

As El Reg goes to press, the Dow Jones Industrial Average is up a mere six-tenths of a per cent, in response to the political unrest in Egypt, and Intel's shares are off only 0.4 per cent. Wall Street has shaken this off.

This Cougar Point SATA I/O flaw does not seem to be as serious as Intel's infamous June 1994 floating-point math flaw, discovered in its Pentium chip and not confirmed to customers until five months later after news of the bug broke in the press. That bug caused Intel a PR nightmare and a $500m charge.

Advanced Micro Devices, which no doubt hope to make hay out of Intel's woes, has had its own chip issues, of course. The quad-core Barcelona Opteron was delayed for six months because of an error in a table lookaside buffer (TLB) on the chips, which could cause machines to halt in certain conditions as they moved data in and out of L2 and L3 cache. Intel released a kernel patch and some BIOS updates to get around the issue, but AMD had no margin of error at the time (in late 2007) because Intel was getting its Xeon act together.

If AMD is tempted to gloat, perhaps it would be better to instead use that energy on an extra-good job of QA testing on its future Bobcat and Bulldozer cores and their processors. ®