Mad Dog 21/21: When Price/Performance Outruns Elasticity
April 5, 2010 Hesh Wiener
Since its inception, the computer business has given customers better value year after year. Customers have responded by buying computers even faster than price/performance has improved. Give more, get more is a nice idea, but in fact the story is not quite that simple. Computing is a harsh competitive environment, ruthlessly Darwinian. The industry has survived many ups and downs, but the same cannot be said of its vendors or architectures. IBM may find that its server business has survived the recession only to be mortally threatened by economic recovery.
IBM has announced a range of Power7 servers for the i and AIX user communities. (These machines support Linux, too, but that doesn’t define their markets.) Mainframes based on the z11 (which will share some components with the Power7 chips and which could end up with a different name) are expected to debut before long. The new Power boxes give customers a lot more bang for the buck than the Power6 generation offered, but that’s not entirely in IBM’s control. IBM has much less pricing power than it used to enjoy. Big Blue has somewhat greater control in the mainframe segment, but comparisons between z boxes and Power family servers make it hard for IBM to price MIPS, memory and most other features the way it wishes. And that is why IBM is about to have a tough time. IBM’s server prices may have to come down farther and faster than sales volume will rise.
There is some price elasticity in the server business, which means that lower prices will stimulate an increase in demand. But the rise in demand in response to lower prices, which is likely to propel the X86 server business to renewed growth and profitability, might not be enough to adequately enrich the Power (and z) segments.
The severity of the problem might not appear right away because every generation of processor hardware seems to get off to brisk start. After the rush of initial orders has faded, however, IBM could be looking at a very gloomy picture in Power country. IBM could be forced to contemplate buying AMD if it wants to remain potent in the processor business. Yes, it’s that serious.
The outline of the structural problem emerges in the IBM i Strategy and Roadmap document previously reviewed here. Ross Mauri, general manager for Power Systems at IBM, explains that last year 90 percent of the i boxes were Power 520 iron, meaning they had one to four Power6 cores.
This year the same kinds of customers will be offered Power 750 machines if they want the latest chips. The Power 750 spans a wider range than the Power 520 but begins with six- and eight-core models. IBM says its entry models, starting at under $35,000, provide more bang for the buck than competitive servers in the same performance range, but the comparisons don’t take into account the new Xeon 5600 and 7500 chips from Intel or their rivals, the Opteron 6100s, from Advanced Micro Devices.
By summer, IBM’s System x division as well as the X64 divisions of Hewlett-Packard, Dell, Oracle, Fujitsu, and others will be scrambling to capitalize on any upturn in server demand that an economic revival might produce. Changes in client technology could add to the pressures on Power, too. Windows 7, still considered a bit risky in enterprise settings, is expected to gain traction in corporate computing by the end of 2010, helping Windows to grab both ends of the corporate networking wire.
Microsoft is driving IBM and Oracle nuts by pricing key software products on a per-socket basis instead of a per-core basis. The result has been a situation in which many users can buy more CPU cores than they absolutely need without paying extra for Windows and its related database system, email server and other middleware.
If IBM wants to offer the same kind of deal to users of its small box Power machines it can’t very well hold back on similar pricing across the entire Power Systems line (although it might be able to maintain a barrier around its System z servers, at least for a while).
IBM really cannot afford to simply retreat from the high volume, middle ticket part of the Power server market. IBM does not appear to have a way to get the kind of revenue from its i and p Power users than it does from their cousins in the z zone. Moreover, it is not clear that IBM’s future in z will be as unfettered as its past has been.
The white paper IBM uses to promote the value of Power servers compared to Windows and Linux alternatives indicates that hardware costs are very similar for all three options and that the i advantage comes from lower personnel costs in the i environment. The argument made against Linux in this study is an argument against Linux on Power as well as on X64 machines. And even though the IBM-sponsored study talks about sites running from 300 to 1,000 seats–a span that is likely to include many ambitious companies–it doesn’t really counter the representation often made by Windows resellers that the Microsoft choice will allow a customer to get more economic headroom than Linux or one of the Power environments because Windows counts sockets rather than cores.
With price/performance rising all the time it’s no wonder that unit sales of servers have to rise considerably just to keep revenue flat. Gartner says that during the fourth quarter of last year server vendors shipped 4.5 percent more units but that aggregate revenue fell by more than 3 percent. While it is possible that this picture has been darkened by a dip in mainframe sales as the z10 reaches the tail end of its marketing cycle and as other buyers waited for Power7 hardware, the story is really pretty much what any observer would expect. Server announcements invariably include details of the improvement in price/performance the current generation offers compared to the past one. There is no such thing as the announcement by a sever vendor that strong demand has enabled the computer maker to increase prices. (That is not true when it comes to electronic components, which do fluctuate up and down in the short run even if long-term trends point toward improved value.)
If IBM wants to increase its server revenue and profitability, it cannot do so by raising prices. While some committed customers might be willing to go along with this–the price per MIPS of mainframes was rising in the 1970s until Amdahl started selling IBM-compatible systems–there simply are not enough buyers to offset the defectors. Plenty of users, probably a majority, would respond to a price increase by looking around for alterative platforms. Even if the alternatives could not at first support legacy workloads, the story would be different for new workloads. The result is that IBM cannot increase prices to compensate for modest sales volumes.
What IBM has to do if it wants Power Systems revenue to rise is to cut prices fast enough to preserve or improve its value proposition compared to that provided by X64 servers. This may eventually be the case with mainframes, too, but for now there are no X64 systems that are trusted to scale up to the capacity of enterprise mainframes. IBM doesn’t seem to be worried about users of small mainframes and may actually be unable to explain why so many of them have hung around for so long.
This is pretty hard to do, even for IBM. Even though Power7 is a leader in the chip world, there is no IBM processor that is sold in the vast volumes characteristic of the X64 family. There may be differences between Intel’s client chips and the CPUs that go into servers, but there is a lot in common, including production processes. Intel has shown that it is not as good at making chips sold in relatively modest volumes as it is when it comes to chips made by the millions. The Itanium suffers from all the problems IBM faces in its Power circuit line and more because Itanium lives in the shadow of siblings that may yet grow to be more powerful and even more versatile.
From time to time there, is speculation in the computer industry about IBM taking over AMD so it can capitalize on the high production volume in the X64 segment. The strongest case against such an acquisition is that the political consequences inside IBM, if Power lost its power, would be unimaginably destructive. But if the Power processor business reached a stage where it was clear it could never make big money on its own, IBM might have to make some hard choices. Even if it’s a loss leader on its own, the chip family is the foundation of IBM’s profitable ventures in Power-related software and services. That gives the Power chip team some security. But if X64 chips eat the bottom half of the Power market during the next three years, which is how long it will probably take IBM to get Power8 out the door, there might not be a Power9. Or if there is, it might not be made by IBM.
The Power7 Systems Sales Pitch
i/OS Gets Short Sheeted with Power7 Thread Counts
The System iWant, 2010 Edition: Clustered Boxes
IBM Fired Up About Power7-Based Smarter Systems
A Little Insight Into the Rest of the Power7 Lineup
Power7: Yields Are Good, Midrange Systems A Go
The Power7 Rollout Begins In The Middle
The System iWant, 2010 Edition: Blade and Cookie Sheet Boxes
The System iWant, 2010 Edition: Entry Boxes
The System iWant, 2010 Edition: Midrange Boxes
IBM Preps Power7 Launch For February
Looks Like i 7.1 Is Coming In April
The System iWant, 2010 Edition: Big Boxes
Power Systems i: The Word From On High
Power Systems i: The Windows Conundrum
Power Systems i: Thinking Inside the Box
Rolling Thunder Rollout for Power7 Processors Next Year
IBM Rolls Up an i 6.1.1 Dot Release
The Curtain Rises a Bit on the Next i OS, Due in 2010
Start Planning for Power7 Iron Now
IBM to Reveal Power7 Secrets at Hot Chips
Power 7: Lots of Cores, Lots of Threads