In the 1990s, it was a good fun to watch the great confrontation between RISC processors (found then mostly on servers and workstations) and CISC processors (primarily x86 chips found in client-side desktops and laptops). RISC servers were massive workhorses, driving enterprise servers as well as most of the major Internet hosts. They could run numerous threads in parallel at blazing speed, they were incredibly reliable, and they enjoyed an architectural purity that made them the darling of pundits, researchers, and academics. They could also run in 64-bit mode, so you could take a mongo RISC box and host a good-sized database all in RAM. At the time, that was a fabulous concept and it made RISC the architecture.
CISC was viewed as a poorly designed, overly complex instruction-set architecture that ran on small machines and struggled to stay in the RISC ballpark. Plus, it barely did parallel. However, CISC had an inexorable and insatiable aspect to it as Intel kept pushing the architecture into new areas. Systems with Intel and AMD processors began eating up the workstation market, especially computationally oriented workstations intended for developers and scientists. When inexpensive GPUs teamed up with x86 processors, the graphics-oriented workstation vendors (especially SGI) were chased up the food chain and ultimately out of the market. SPARC and HP workstations followed next. And then, as servers migrated from big boxes to clusters of inexpensive x86 boxes, the rout of RISC architectures became almost complete. MIPS is now mostly embedded, HP's PA-RISC is gone, SPARC lives on in a fraction of Oracle's servers. IBM Power is perhaps the largest unconquered architecture: It's very much alive in Big Blue's machines and in gaming consoles.
(To be fair, though, RISC did score an architectural triumph in the heart of CISC. Intel x86 processors use a RISC-based instruction execution unit. CISC instructions are converted to an internal RISC format before being executed.)
A computing market with CISC dominating PCs and servers, and RISC occupying small niches and fringes has been the lay of the land for the last five to ten years. But now comes a new RISC competitor for x86 that enters the arena touting the same low cost and wide adoption on small devices that Intel and AMD used 15 years ago.
That architecture is ARM, the RISC-based silicon found in most mobile phones today. ARM brings many interesting benefits. The first of which is a huge installed market. At one time, ARM held a 98% market share in phones. It still remains the leading chip in phones, and it leads media players, and other consumer devices. The licensing company, ARM holdings, received royalties on more than two-billion ARM processors last year. That's a lot of chips and a lot of existing software.
What makes ARM attractive to vendors is its low energy consumption, simple RISC architecture, and large installed base. To truly challenge x86, the architecture must make the leap from the small devices it favors now to PCs and eventually servers.
That transition appears to be beginning. Earlier this year, Microsoft announced that Windows 8 will ship on two hardware platforms: x86 and ARM. That is a substantial endorsement. In high-performance computing, ARM is showing up as well. At Supercomputing '11 in Seattle this week, Nvidia announced new hardware boards for CPU/GPU computing in which the CPU is ARM, not x86. The new hardware, code-named "Mont Blanc," is driven by the observation that, "In most current [CPU/GPU] systems, CPUs alone consume the lion's share of the energy, often 40% or more. By comparison, the Mont Blanc architecture will rely on energy-efficient compute accelerators and ARM processors used in embedded and mobile devices to achieve a 4x-to10x increase in energy efficiency by 2014," according to Alex Ramirez of Nvidia. That's a substantial benefit.
Intel has known about the threat of low-power, high-performance CPUs for a long time; and it tried to enter that space with the Intel Atom processor, made up of one or two stripped down x86 cores. Atom chips were a success in the netbook market, but were not power-efficient enough to be able to move more deeply into the regions dominated by ARM; namely, smaller power-sensitive devices.
As the Atom demonstrates, the IA-32 architecture is not easy to implement at low power levels with good performance. So, ARM may find its path upwards fairly open for the time being. However, Intel has faced challenges before and prevailed. Its extraordinary lead in production and fabrication, plus its longtime commitment to research, make it a formidable player that won't stand around long watching an upstart threatening to take market share.
Either way, this is going to be a fascinating battle to watch unfold.