BY TOM WILLIAMS, EDITOR-IN-CHIEF
Intel must be feeling like a bear being circled by wolves. The wolves are named ARM and they are gradually tearing meaty chunks out of the x86 bear. But beware of getting a bear cornered. A cornered bear can lash out with some pretty mean claws as well. For some time there has been a battle going on in the embedded space between the Intel Atom family of x86 low-power processors and a variety of ARM architecture-based devices, which are increasingly centered on the ARM Cortex M and R series cores. Intel’s advantage has been that the embedded world has largely been derived from that of the PC—high-volume, low-cost devices as well as widely used peripheral technologies like USB and PCI and now PCI Express. ARM’s big selling point has been its low power consumption, which despite all its efforts, Intel has never been able to match. This difference is being increasingly exacerbated with the surge in mobile devices, wireless connectivity and the need to save power at every step. This and the flattening of the PC market has allowed ARM and its many licensees to make steady gains. Add to this mix, Intel’s old nemesis, AMD, which has successfully marketed lines of processors based on its clean room-developed version of the x86 instruction set. Now even AMD is starting to move into the ARM arena.
It appears that so far Intel’s strategy has been to more aggressively and specifically target the embedded arena with such things as Atom and Core-based SoCs that incorporate a range of on-chip peripherals and increasingly to move into the server arena with powerful versions of its multi-core Xeon processors. Unfortunately, the lust for power-savings is also now rampant in server farms, which consume enormous amounts of electricity and dissipate ungodly amounts of heat—all of which can be visualized as dollar signs by their managers. And now ARM is targeting the server arena with a new line of 64-bit ARM cores and AMD, among others, have started offering processors targeting the server market based on these new cores.
Another thing that Intel has to contend with is the emergence of heterogeneous system architectures (HSAs) that integrate such elements as graphic processors, DSPs and FPGAs on the same die. This has led to the emergence of the HSA Foundation, among whose goals are the establishment of standards and open-source software tools to address the combination of CPUs with parallel elements on the same chip. Two of HSA’s founding members are ARM and AMD. But don’t count Intel out yet. The bear may be cornered, but it’s getting ready to strike.
There are rumors floating around the industry—and by the time of publication they may be more than rumors—that Intel is in talks to purchase Altera, a major producer of FPGAs. If true, it makes sense on so many levels. Intel and ARM have fundamentally different business models. Intel has its in-house technology, which it will not license to anyone—the bear. ARM manufactures nothing itself but licenses its IP out to a host of different manufacturers—the wolves. To compete, Intel needs the in-house capability to integrate its processor technology, and all its legacy software, to programmable logic on-chip.
Both Altera and its major rival Xilinx currently offer families of CPU/FPGA hybrid processors on which the CPU is an ARM device. For Xilinx, these are the Zynq products and for Altera they are the SoC FPGA products. No FPGA company will be able to license something like Atom or Core i5 IP from Intel to build a competing device. So if Intel does acquire Altera, you can expect it to, of course, continue to support and develop Altera’s technologies and serve its customers. There were predictions that after its acquisition of Wind River, it would drop lines that served other processors, but that did not happen. We can, however, expect to see a discontinuation of ARM-based SoC FPGAs for a line of hybrids based on Atom and Core architectures.
Originally, it looked like the rivals were going to duke it out in the embedded arena and that is certainly happening. Now, however, with products like AMD’s Opteron A1100, which is based on the 4- or 8-core 64-bit ARM A57 core and Xilinx with a device based on a quad-core Cortex-A53, attention is clearly turning to the server market as well. Currently, AMD’s parallel processing element is the multicore Radeon GPU architecture it acquired from ATI, which also lends itself very well to intensive numeric processing.
If Intel does acquire a major programmable logic house like Altera with its built-in expertise, it will be able to turn out hybrid processors with a huge range of specific (network server) functions as well as generally programmable and configurable devices. An added advantage is that FPGAs consume less power than the Xeon devices and while not as low as ARM, can make a good case in terms of compatibility with legacy software. So far programmable logic companies have licensed CPU (ARM) technology. They must therefore make exact decisions as to which core(s) to license. There could be an interesting advantage when a company with complete control over a vast family of CPUs and specific devices with combinations of on-chip peripherals and I/O also actually owns a programmable logic company. The possibilities could be endless . . . and as powerful as the claws of a bear.