Abstract:
This follows a data processing system comprising multiple GPUs (2, 4, 6, 8) includes instruction queue circuitry (28) storing data specifying program instructions for threads awaiting issue for execution. Instruction characterisation circuitry (30) determines one or more characteristics of the program instructions awaiting issue within the instructional queue circuitry (28) and supplies this to operating parameter control circuitry (20). The operating parameter control circuitry (20) alters one or more operating parameters of the system in response to the one or more characteristics of the program instructions awaiting issue.
Abstract:
In described examples, an energy efficient task scheduler for use with a processor provides multiple reduced energy use modes. In one embodiment, a system (100) for executing tasks includes a processor (102) and a task scheduler (106). The processor (102) provides multiple different reduced energy use modes. The task scheduler (106) is executable by the processor (102) to schedule execution of multiple sleep tasks (116). Each of the sleep tasks (116) corresponds to a different one of the reduced energy use modes. The task scheduler (106) is executable by the processor (102) to execute each of the sleep tasks (116), and as part of the execution of the sleep task (116) to: place the processor (102) in the reduced energy use mode corresponding to the sleep task (116), and exit the corresponding reduced energy use mode at suspension of the sleep task (116).
Abstract:
Computing devices receive power from multiple fuel cells, consuming natural gas and outputting electrical energy natively consumable by the computing devices. The fuel cells are sized to provide power to a set of computing devices, such as a rack thereof. The computing devices of a failed fuel cell can receive power from adjacent fuel cells. Additionally, the fuel cells and computing devices are positioned to realize thermal symbiotic efficiencies. Controllers instruct the computing devices to deactivate or throttle down power consuming functions during instances where the power consumption demand is increasing faster than the power being sourced by fuel cells, and instruct the computing devices to activate or throttle up power consuming functions during instances where the power consumption demand is decreasing faster than the power being sourced by the fuel cells. Supplemental power sources, supplementing the fuel cells' inability to quickly change power output, are not required.
Abstract:
Techniques described herein generally include methods and systems related to the use of processors that include graphene-containing computing elements while minimizing or otherwise reducing the effects of high leakage energy associated with graphene computing elements. Furthermore, embodiments of the present disclosure provide systems and methods for scheduling instructions for processing by a chip multiprocessor that includes graphene-containing computing elements arranged in multiple processor groups.
Abstract:
A wrist-worn athletic performance monitoring system, including a gesture recognition processor configured to execute gesture recognition processes. Interaction with the performance monitoring system may be based, at least in part, on gestures performed by the user, and offer an alternative to making selections on the performance monitoring system using physical buttons, which may be cumbersome and/or inconvenient to use while performing an athletic activities. Additionally, recognized gestures may be used to select one or more operational modes for the athletic performance monitoring systems such that a reduction in power consumption may be achieved.
Abstract:
In one embodiment, a method includes determining a power consumption profile for a device. Status information for the device may be received, wherein the status information comprises power status and network connectivity status. Using a resource-control algorithm and based on the status information and the power consumption profile, a schedule for sending push events to the device may be determined. Content to be provisioned to the device may be identified, and the scheduled push events may be sent to the device, in order to provision the identified content to the device. The resource-control algorithm may be further based on one or more device-based consumption factors, such as a periodic data usage transfer limit with respect to a specified network and a data usage status with respect to the specified network, or system-wide consumption factors, such as a power threshold that applies across all devices.