Intel's ISSCC presentations include a chip capable of running at just 280mV and a new variable-precision floating-point engine.
Intel has announced development of an ulta-low voltage processor which is capable of operating at voltages as low as 280mV, claiming the feat will help it create future chips with a five-fold increase in energy efficiency.
Announced ahead of the International Solid-State Circuits Conference in San Francisco, the chip is unlikely to win any performance awards. Based on a 32nm implementation of the original Pentium architecture, the prototype chip boasts a dynamic range reaching from 3MHz to 915MHz.
The result is the Intel Near Voltage Threshold Processor, which can tick over at a low speed while drawing just 280mV. When in 'turbo' mode, at these speeds something of a relative concept considering the top speed of the NVT process is under a gigahertz, the chip runs on just 1.2V.
Intel also announced a new design method for floating-point processing, which allows the prototype processor to vary the precision of its calculations dynamically. The result, Intel claims, is a boost in energy efficiency of around sevenfold over traditional implementations. According to Intel's planned presentation, the variable-precision floating point fused-multiply add unit hits 52 to 162 gigaflops per watt at 1.45GHz. The technology will be used in Intel's future graphics products, the company confirmed, along with a vertex and pixel shading lighting accelerator capable of processing 2.05 gigavertices a second while drawing just 151mW.
The announcements come as Intel looks to lead the way into 'exascale' computing: processing at an exaflop level, rather than the gigaflop or teraflop levels of current systems. To reach that goal, the company is attempting to create processor designs that boast increases in energy efficiency of between fifty and a hundred times that of current chips.
Much of the company's technology in this regard is limited to the supercomputing market. The Many Integrated Cores (MIC) cards, Knights Ferry and Knights Bridge, as an example are fifty-core highly-parallel processors that are unlikely to find their way into the average gaming rig.
While the technology is being developed for supercomputing applications, however, it will trickle down to the consumer level over time. Intel has already declared war on ARM in the tablet and smartphone space with a system-on-chip (SoC) variant of its Atom processor family with in-built Wi-Fi connectivity. Building in a five-fold increase in energy efficiency would certainly help encourage device makers to make the leap from the ARM instruction set architecture to x86.
Intel's work also holds the promise of cooler, faster chips on the desktop. As the number of transistors on a processor increases, so too does the risk of current leakage and heat dissipation issues. A vast increase in energy efficiency will certainly help the company in those regards until a practical alternative to silicon can be found and proved to work on a commercial scale.
Sadly, the company has yet to offer a hint as to when the technologies shown of at ISSCC will be making their way to the production lines, instead preferring to talk up its Ivy Bridge processor family featuring tri-gate transistor technology.