Intel's president and CEO Paul Otellini said in a conference call with Wall Street analysts after the market closed yesterday that the "Sandy Bridge-EP" Xeon E5 processors and their related "Romley" server platforms, are now in volume shipment and due to be launched during the first quarter, as was widely speculated.
The Xeon E5s, their "Patsburg" C600 chipset, and related motherboards were expected to be announced in late Q3 and then maybe in Q4 – but that didn't happen for reasons that Intel did not divulge. The speculation is that the C600 chipset has a bug in one of its SAS controllers, similar to one in the PC variants of the C200 chipsets that delayed the launch of desktop and entry server Sandy Bridge chips this time last year.
Intel never disclosed publicly that the Xeon E5s were late, and is sticking by its position that they are therefore not late, and moreover was bragging last September that hyperscale cloud and supercomputer customers were getting Xeon E5s on an NDA basis and in volumes sufficient to make them contribute to Intel's financials.
Intel showed off its Romley motherboards at the end of November last year and has over 22 different SKUs of servers and boards ready to go.
Now we know that the Xeon E5 launch is in the first quarter, not just "early 2012" as Intel was saying late last year, with new 22-nanometer "Ivy Bridge" chips for PCs coming in "early spring", and thus setting up the server line for a whole new line of Ivy Bridge upgrades.
As El Reg previously reported, Intel posted $13.9bn in sales in the fourth quarter of 2011, up 22 per cent, with net income of $3.4bn, up only 6 per cent. For the full year, Intel's sales rose by 24 per cent, to $54bn, and it scratched $12.9bn on the bottom line in black ink, 13 per cent more than in 2010.
Intel would no doubt have liked for profits to grow faster than revenues in Q4 and for the full year, but that is not possible with a PC slowdown and a delayed server ramp. That said, this was a record year for Intel and nothing to be ashamed about, and Sandy Bridge processors and chipsets across PCs and servers accounted for 40 per cent of total sales.
Intel's Data Center Group, which makes chips, chipsets, and mobos for servers and workstations as well as networking and communication gear, grew a bit slower than Intel overall, with sales only up 8 per cent to $2.72bn in the fourth quarter. Operating income was up 1.9 per cent to $1.45bn. For the full year, Data Center Group had $10.1bn in sales, up 17 per cent from a year ago, and operating income for the twelve months was $5.1bn, up 16.2 per cent from 2010.
It's clear that profits were under pressure in Data Center Group in Q4, no doubt because of the Xeon E5 delays. Still, on a quarterly and annual basis, Intel's server-related sales are rising faster than the server market overall, which means its Xeon platform is more of the server pie. And it's not like AMD is mounting a huge threat to Chipzilla these days.
Intel took ten years to get Data Center Group to $10bn in annual sales, and general manager Kirk Skaugen – who became the new head of the PC Client Group on Friday morning in a management shakeup – thinks by expanding Intel's presence in storage and networking, the group can push up to $20bn in the next five years.
In the conference call with Wall Street, Intel CFO Stacey Smith said that Intel would therefore be making investments in the server racket and its related storage and networking areas in the data center. Smith added that Intel did not see "softness" in its server business, if you look at it over a longer cycle than a 2010 to 2011 compare.
This is totally silly.
Yes, Intel's server-related business was up smartly at the end of 2010, and yes, that is a tough compare. But the business also went into a black hole in the years before that, so the compares were easier. Had Intel actually launched the Xeon E5s in September at its own Intel Developer Forum and been able to deliver them to server makers throughout the fourth quarter, the company would have very likely have had a server business that grew faster than Intel overall and delivered more profits.
The main thing that Otellini wanted to get across is that Intel customers and investors had better get used to this choppiness and lumpiness, and not just because of product-transition blimps and bloopers that happen from time to time, but because the nature of the server business is changing.
"The data center business we have today is not your grandmother's server business that we had for many years," Otellini explained. Intel Xeons and SoCs are in storage and networking products now, and they are on different cycles than the server business.
Moreover, hyperscale data centers skew sales. "These sales are lumpy," said Otellini. "They tend to be a function of when Facebook, Google, or Amazon decides to turn on a new data center and buy X-hundred-thousand units. Or there is a new generation and they want to have a quick complete swap-out. And as a result, we're seeing a change to the historical linearity that we saw in the enterprise data center business." He advised Wall Street to get used to "a little more lumpiness."
In a sense, the server OEMs could buy at a near steady state, and then their own reseller channels would be shock absorbers, mitigating against the ups and downs of the economy and state of the millions of server buyers worldwide, and taking machines in a nearly continuous stream that suited server manufacturers more than resellers and perhaps potential customers. Cloud providers seem to be a little less generous with keeping computing capacity in inventory that is not making money, and they buy in big blocks for large numbers of customers.
This is perhaps just a preview of what your granddaughter's server business will look like. ®