A widely reported story at TechSpot that a flaw in the forthcoming 2013 generation of Core microprocessors for PCs and notebooks (codename Haswell) led to speculation that Intel would delay some but not all Haswell chips until a fix in silicon could be made subsequent to a June launch. Now all of this is unconfirmed by Intel, so take the following analysis in this blog post with a grain of salt.
The reason I’m writing anything is that Intel can ill afford a delay in its mainstream processor delivery schedule for clear business reasons that I’ll outline below. And waiting on another stepping of Haswell would clearly cross a dateline that leads to all sorts of follow-on inefficiencies ad infinitum. However, a stutter-step in production timing might get everything back on track in 2014.
The Ideal PC Annual Launch Schedule
The PC industry has a cadence which Intel has fine-tuned with its microprocessor launch and production schedule. My stylized version looks like this:
- Year – 1: all the testing and sharing with partners for design purposes
- Year – 1, December: begin shipping revenue units for partner’s early production and demo units. This looks good when reporting Q4 results to Wall Street in mid-January.
- January: launch the years’s generation of technology at, say, the Consumer Electronics Show. This gets loads of press attention and helps to freeze technology purchases until the products ship. January automobile shows serve the same purpose.
- February and March: partners begin shipping hero products aimed at enthusiasts and thought-leaders, but not in high volumes. Also, get evaluation units into the hands of IT departments. The chip fabs are ramping efficiency now.
- April and May: IT departments select the year’s PC and notebook standards and begin an annual refresh cycle by July.
- By June: mainstream chip launch and the partner’s fall product roll-out. Collect orders from retailers.
- July and August: Asia makes volume PCs and notebooks and ships them into retail.
- August – September: Back-to-school is a major consumer refresh period, where the new products are on display and students, in particular, look for retail new tech purchases.
- October: Asia PC manufacturers gear up for the holiday selling season. By now, the Intel fabs are running like clock-work, at high yields. The third major chip product launch of the year, typically Xeons for servers (but it could be Atoms for mobile devices in the future) occurs now.
- November – December: Holiday tech purchases is the largest volume selling period of the year. And repeat the revenue shipment for the next year’s products before January.
We’re Not on the Ideal Launch Schedule
Haswell has not been launched yet, and probably won’t until June, not January 2013. How did this happen? That’s not the topic so I’ll spare you the drawn out details. The synopsis: Sandy Bridge in 2011 had a chip set flaw, which halted shipments in January and slowed that year’s roll-out by a quarter, so that in January 2012 the industry was not ready (e.g., had inventory and equipment to write off) for Ivy Bridge. In late 2012, a PC industry slow-down left everybody with inventory that they wanted to burn off before starting with new kit for Haswell, which requires a new motherboard not compatible with Sandy or Ivy Bridge. So that’s how we got here.
Refer back to the ideal launch schedule. By June 30th, the industry needs to have introduced all of its performance and mainstream products, taken orders, and contracted Asia to produce the tens of millions of units needed for a successful back-to-school and holiday selling period.
That’s not going to happen in 2013. A big bang of products and production by the PC industry starting in June to catch up with August back-to-school is too risky a business move to contemplate. The industry would not move as one. Problems and finger-pointing. On top of that scenario, throw on the blog lead with a possible Haswell chip flaw that, at minimum, Intel wants fixed but may not delay the Haswell launch for the fix.
So, what’s the second half outlook for the PC industry?
Status Quo or Stutter Step?
Here are the three scenarios I see for the second half:
1. Launch Haswell in June with a big bang. Try to get the new products into retail by August in depth and in volume by compressing the introduction and production calendar. This is the least likely and highest risk scenario.
2. Launch Haswell in June. Pretend June is January on the ideal calendar, and ramp Haswell in the second half alongside refreshed Ivy Bridge (i.e., Third Generation Core) products. Creates consumer confusion with two generations on sale at retail at once, but that’s been the reality anyway. Leaves the merchandising problem to the retail channel. This scenario is the low-risk, low reward bet.
3. A Stutter Step in marching is a quick step or skip that gets an out-of-step marcher back in cadence. In this scenario, Intel would ramp Haswell in the second half alongside refreshed Ivy Bridge as in Scenario 2. Haswell is the Tick in Intel’s Tock-Tock cadence, with a tick being a new architecture and the tock a shrink of the architecture. The stutter-step in January 2014 would launch the 14 nm shrunk version of Haswell, code-name Broadwell:
- Bringing on Broadwell in January would get Intel and the industry back on the ideal schedule. The assumption is that Intel is ready for 14 nm, and I have heard nothing to dissuade me from that assumption.
- The ideal schedule is the lowest annual risk and the highest profits for the industry.
- The status quo is a time-bomb waiting for the next flaw. As discussed above, June is already a crap shoot for a big bang launch into the holiday season. Any future delay (and I think the complexity of today’s chips favors flaws and delays) would likely overlap the biggest selling season with the newest technology launch. Ouch, cognitive buyer’s dissonance.
- Since Haswell and Broadwell are interchangeable at the motherboard level, risk is mitigated because a 14 nm fab hiccup could be remediated with Haswell chips.
- Any move to speed up the annual cadence adds risks, and the stutter-step scenario cuts the economic life of Haswell to six months-plus, not twelve-plus. However, 14 nm in volume is double-digits percentage less expensive to manufacture, lowering chip costs and improving margins. Gentlemen, start your net-present-value spreadsheets.
- Delaying 14 nm production until mid-2014 puts a huge crimp in Intel’s mobile strategy, which desperately needs to get ahead of the technology competition the sooner the better.
- There are plenty of remaining uses for the idle 22 nm fab equipment, including Ivy Bridge, remainder Haswell, and 22 nm Xeon and Atom products yet to launch in 2013.
Send your comments to Twitter: @peterskastner