“The enemy of my enemy is my friend.” That reasoning has led to many strange alliances among nations, but it also makes for some interesting business partnerships. For example, at AMD’s recent Fusion developers’ conference, AMD and ARM were practically flaunting their PDAs—public displays of affection, not personal digital assistants.
ARM is the world’s leading supplier of 32-bit CPU cores for cell-phone processors, consumer electronics, and embedded systems. Lately, this relatively small British company has been irritating Intel, which is trying to muscle into ARM’s low-power territory, so AMD and ARM share a common enemy. Their hugfest at AMD’s conference has important implications.
AMD is panting over heterogeneous microprocessors, which integrate multiple processor cores of different architectures on the same chip. AMD’s Fusion processors are heterogeneous, because they combine AMD’s x86 CPU cores with an AMD Radeon GPU core. More particularly, AMD is lusting for heterogeneous processing—sharing general-purpose workloads among different cores, instead of dedicating some cores to specialized tasks.
Programmers are already using GPUs for general-purpose computing on AMD’s ATI Stream and Nvidia’s CUDA software platforms. Before Fusion, however, the GPU was a second-class citizen, relegated to the PCI Express bus. Fusion couples the GPU directly to the CPUs, making them peer processors. Now AMD is suggesting that ARM’s CPU cores could make it a threesome.
ARM cores could appear in PC processors or in future AMD embedded processors. They could work alongside the x86 CPUs and GPU, offloading some chores that require better power efficiency. Programming becomes more complex, but the chip could save power and cost less. This union could also save AMD the trouble of developing very-low-power x86 cores.
If the flirting between AMD and ARM turns serious, maybe Intel will be the one who’s hot and bothered.
Tom Halfhill was formerly a senior editor for Byte magazine and is now an analyst for Microprocessor Report.