Thousand-core chips; data-driven physics; carbon-footprint-tracking wearable device.
Cache-coherence innovation for thousand-core chips
MIT researchers are getting ready to unveil what they say is the first fundamentally new approach to cache coherence in more than three decades. They reminded that in a modern, multicore chip, every processor core has its own small memory cache, where it stores frequently used data. The chip also has a larger, shared cache, which all the cores can access. If one core tries to update data in the shared cache, other cores working on the same data need to know, so the shared cache keeps a directory of which cores have copies of which data.
However, that directory takes up a significant chunk of memory. For example, in a 64-core chip, that directory might comprise 12 percent of the shared cache, which will only increase with the core count. Envisioned chips with 128, 256, or even 1,000 cores will need a more efficient way of maintaining cache coherence.
In existing techniques, the directory’s memory allotment increases in direct proportion to the number of cores, but with the new approach — to be detailed at the International Conference on Parallel Architectures and Compilation Techniques next month — the memory allotment increases according to the logarithm of the number of cores. In a 128-core chip, this means that the new technique would require only one-third as much memory as its predecessor, and with a 256-core chip, the space savings rises to 80 percent, and with a 1,000-core chip, 96 percent.
A new way of computing under development by University of Michigan researchers could lead to immediate advances in aerodynamics, climate science, cosmology, materials science and cardiovascular research.
To this end, the National Science Foundation is set to provide $2.42 million to develop a facility for refining complex, physics-based computer models with big data techniques at the University of Michigan, with the university providing an additional $1.04 million.
The focus of the project is a computing resource, called ConFlux, designed to allow supercomputer simulations to interface with large datasets while running, which is expected to close a gap in the U.S. research computing infrastructure, the researchers said, and put them at the forefront of the emerging field of data-driven physics.
Specifically, the project adds supercomputing nodes designed specifically to allow data-intensive operations; the nodes will be equipped with next-generation CPUs and GPUs, large memories and ultra-fast interconnects. A three-petabyte hard drive will handle both traditional and big data storage.
Tracking carbon footprint with wearable device
A wearable technology developed at by University of Washington researchers senses what devices and vehicles a user interacts with throughout the day, which helps track that individual’s carbon footprint, enable smart home applications or even assist with elder care.
According to the researchers, in today’s smart home, technologies can track how much energy a particular appliance like a refrigerator or television or hair dryer is consuming, but they don’t typically show which person in the house actually flicked the switch.
The sensor worn on the wrist uses unique electromagnetic radiation signatures generated by electrical components or motors in those devices to pinpoint when its wearer flicks a light switch, turns on a stove or even boards a train. At the end of a day or month, the user can see how much energy they’ve used.
The researchers said the MagnifiSense device has potential for other smart home applications, such as recognizing a user’s preference for interacting with an appliance or device. By sensing whether an adult or child is turning on a television or tablet, for instance, a system could automatically display their favorite programs or tailor the device with appropriate selections. Or, in assisted living settings or nursing homes, the wearable sensor could help keep track of how efficiently elderly people are going about everyday tasks such as cooking or grooming. It could also detect when a stove has been left on for a long period of time and help alert someone to that danger.
The sensors also capture a broad frequency range that allows the system to differentiate between electromagnetic radiation emanating from the unique combinations of electronic components such as motors, rectifiers and modulators embedded in everyday devices.
Further, the team developed signal processing and machine learning algorithms to help the system correctly match those patterns with a particular type of device.
One advantage to a wearable option is that anyone concerned about privacy issues can control when they use it, researchers said, or simply take it off.