Mad Dog 21/21: ARMs To Fare Well
November 7, 2011 Hesh Wiener
Steve Jobs was such a captivating promoter of inventions that his products reshaped our thinking, defining or redefining products we once thought we fully understood. At his best, Jobs was almost too good. If Picasso were God all fish would be flounders. But the computer industry, like nature, fosters diversity. Apple‘s smart clients, the iPhone and iPad, are iconic devices built around systems-on-a-chip (SoCs), but they are not the only important applications of this technology. Servers, too, can be made from compact, efficient, and inexpensive SoCs. And they will prove to be exceedingly disruptive.
Two examples that have been around for a year or two have been showcase developments for Marvell, the semiconductor company that bought Intel’s ARM architecture processor business in 2006. Marvell donated a handful of circuit boards to a group that ported Linux to ARM and hosts its website on a small cluster of ARM blades that were built about 18 months ago. The other Marvell showcase application is a family of tiny home servers from Tonido that is now in its second generation. The original Tonido product, TonidoPlug, was the size of a phone charger; the current one is about the size of a sandwich. It’s larger because it has space in it for a 2.5-inch hard drive in case users want to store very large libraries of personal files. Without the disk the server, including an Ubuntu Linux system, a stack of applications including an Apache web server, a WiFi hub, and USB port sells for under $200. Basically, Tonido or some other gizmo like it provides a nice home base for an iPad or rival tablet that is also reached via the Internet through a service supplied by Tonido.
Marvell is hardly the only ARM licensee that would like its customers to get into the computing mainstream. Nvidia, famous for its graphics chips, has a five-ARM-engine chipset that powers the Asus Transformer 2, a tablet with a snap-on keyboard. Asus, which invented the netbook type of computer a few years ago, hopes to reclaim territory it has lost in part to the iPad and its ilk, in part to the latest ultralight notebook computers such as the MacBook Air. The Transformer isn’t a server; it’s a tablet/notebook with very powerful graphics engines. But just as Intel Xeon workstations and servers are close cousins, the opportunity for Asus to market small servers that recycle the engineering behind the Transformer is obviously right at hand. Moreover, the company that Asus is working with on high-end ARM-based client devices, Nvidia, is one of a handful of firms that have already licensed the 64-bit server chip core, called ARMv8, that ARM Holdings plans to offer as the next major generation of ARM processors.
IBM is well aware of developments in the ARM world and happens to be one of the many companies that own some kind of ARM license, which means that Big Blue can make chips based on ARM architecture. For all we know, ARM may be used inside IBM server channels or other components already. But that doesn’t mean IBM is really ready for the kind of move ARM and the other SoC architecture, MIPS, may force Intel to make.
What Intel may have to do, even as it strives to create ultra low power versions of its Atom chips (or perhaps an entirely new family of X86-X64 chips for SoC applications), is to move upstream drastically faster. Intel cannot hold back on the rise of X64 to protect its Itanium circuits while its low-end, high volume market is under threat and its lucrative server chip family suffers continued assaults from X64 rival Advanced Micro Devices. Basically, Intel has to do its best to clobber Power and mainframes grab as much server market share as it can.
IBM, on the other side of the market division, has to revisit its past choices, decisions that involved moving upstream to faster and faster chips while largely abandoning efforts to make circuits that might compete in the lower-performance, higher-volume part of the server market. Just as IBM left (or was forced out of) the client chip business, its Waterloo the decision by Apple to use Intel chips in Macs, IBM may now be forced to reconsider how to keep a grip on entry level servers for the AIX and IBM i crowds. IBM might not be able to do this with Power chips because the unit volume simply is not there. Marketing considerations militate against IBM’s use of Intel technology to support its AIX and IBM i software, although if any company could manage the kind of software migration this would require, IBM is the one to do the job. That leaves IBM with two other options in the form of RISC chips that are cheap, powerful, and part of markets with huge unit volumes and the kind of software and hardware tools that could meet IBM’s needs: ARM and MIPS.
ARM, according to its financial statements, expects its licensees to make about 8 billion processor or SoC chips this year. That’s right, 8 billion. MIPS chips are made by the tens of millions and yield licensing revenue of $80 to $100 million a year. The MIPS architecture has long since included 64-bit processing and virtualization hardware but the circuit’s success in set-top boxes, data routers, and other market segments has never extended to the phone business. Still, if it’s servers that are the point, MIPS may not enjoy the glamour or market size of ARM but it does offer established chip designs that could support AIX or IBM i systems. Either ARM or MIPS technology, then, could give IBM a low-cost alternative to Power with which to push back as Intel tries to gain ground in glass house country.
Moreover, particularly if it decides to go with ARM rather than MIPS, IBM can take advantage of the work done by some of Apple’s rivals, work that has pushed down the cost and pushed up the volume of SoCs. The current obvious example of a rising SoC supplier is Texas Instruments, believed to be the source of designs if not actual circuits that will be going into the product most observers consider iPad’s most serious rival, the Kindle Fire. (Bloomberg News says the company making chips for the Kindle Fire is United Microelectronics, but has not said whether the United circuit is, as widely believed, a TI design. TI is considered a big player in the e-reader market because its ARM-based SoCs have superior security technology that can be crucial in the management of copyrighted material used in e-books.)
However Intel and IBM react to the emerging presence of ARM (and possibly MIPS) SoCs at the low power and low power consumption end of the server business, it is obvious that price is going to be one of the major and most deadly weapons used by incipient server makers. Nobody is going to buy an SoC based server because it’s cute to look at. But if it’s cute on the budget, well, that’s another story entirely.
Meanwhile, as IBM is, we suspect, contemplating the future directions of the server market, its larger rival Hewlett-Packard and its smaller competitor Dell have got to be working overtime on their possible entries in the ARM (or maybe MIPS) server game. HP is working with Calxeda on a hot (but cool running) server family that will have a zillion ARM cores. Calxeda is a Texas outfit that got part of its seed money from Texas Instruments, so you can see that there is some serious brainpower and money behind it, whether or not HP turns out to be its best customer. Dell is based in Texas and that increases the odds of its engineers and execs knowing Calxeda folk but Dell could easily reach out to any company, near or far, if it wants outside help in the SoC server race.
But HP and Dell might not be IBM’s biggest rivals even if they add SoC servers to their product lines well before IBM can come up with countermeasures. No, IBM’s biggest challenge might turn out to be its former PC maker, Lenovo, which would love to become a power in the server business. The disruptions that could arise as SoC servers come into the market could provide an opportunity for a new player to reach the portal of the glass house. However, it is also possible, in this perpetually surprising universe, for Lenovo to become not an adversary of IBM but rather its supplier and savior.
The least likely possibility seems to be for IBM to do essentially nothing and still retain its leadership in the server business. IBM’s big-engine, power-hungry servers are like the 14-inch disks Big Blue once made when it ruled the storage business. Science and market forces are pushing the server business toward machines built from swarms of tiny circuits, machines that are more like the systems powering Google and Amazon and the Apple iCloud than the ones whumping out charge card statements. The giant server farms may currently be powered mainly by Intel or AMD chips, but they could well become the first major homes of SoC ARM or MIPS servers. They are not going to become part of the Power base. And, we reckon, as the outfits with the big market capitalization go, so, too, like it or not, will the rest of us.