Alibaba Group Holding Limited

Lingling Jin

Ying Chu Wang

Yijun Lu

Bo Yang

Yi Liu

Wei Zhang

Ling Xu

Qinggang Zhou

Hong Tang

Jun Song

Lingjie Xu

Hong Zhou

Zhengxiong Tian

Li Zhao

Wei Wei

Lin Yu

Jiangwei Huang

Guan Wang

Shuiwang Liu

Nan Zhang

Jun Ruan

Yu Zhou

Shiwei Li

Chao Qian

Xiaolin Meng

Ying Wang

A method for power management based on synthetic machine learning benchmarks, including generating a record of synthetic machine learning benchmarks for synthetic machine learning models that are obtained by changing machine learning network topology parameters, receiving hardware information from a client device executing a machine learning program or preparing to execute a machine learning program, selecting a synthetic machine learning benchmark based on the correlation of the hardware information with the synthetic machine learning models, and determining work schedules based on the selected synthetic machine learning benchmark.

Methods and devices for power management based on synthetic machine learning benchmarks

Systems and methods for a hybrid system-on-chip usable for predicting performance and power requirements of a host server include a big cores module, including central processing units, for receiving and pre-processing performance and power metrics data of the host server and to allocate computing resources, a small cores module, including massively parallel processing units, for mapping each instance associated with host server in the performance and power metrics data to a corresponding massively parallel processing unit based on the allocated computing resources for a per-instance metrics calculation, and an artificial intelligence (AI) accelerator for calculating performance and power prediction results based on the per-instance calculations from the small cores module.

Hybrid system-on-chip for power and performance prediction and control

Embodiments of the present disclosure provides systems and methods for batch accessing. The system includes a plurality of buffers configured to store data; a plurality of processor cores that each have a corresponding buffer of the plurality of buffers; a buffer controller configured to generate instructions for performing a plurality of buffer transactions on at least some buffers of the plurality of buffers; and a plurality of data managers communicatively coupled to the buffer controller, each data manager is coupled to a corresponding buffer of the plurality of buffers and configured to execute a request for a buffer transaction at the corresponding buffer according to an instruction received from the buffer controller.

System and method for batch accessing

The present disclosure provides a system and a method for power management of accelerators using interconnect configuration. The method comprises receiving a power management command comprising a power budget and a designated accelerator, identifying a port associated with the designated accelerator from a plurality of ports of an interconnect, determining a target data transmission parameter for the designated accelerator according to the power budget, and controlling a data transmission parameter relative to the target data transmission parameter through the port associated with the designated accelerator. The present disclosure further provides a non-transitory computer-readable medium that stores a set of instructions that are executable by one or more processors of an apparatus to perform the method for power management of accelerators using interconnect configuration.

Power management system and method implementing interconnect controls

The present disclosure provides methods and systems directed to providing quality of service to cluster of accelerators. The system can include a root connector; an interconnect switch communicatively coupled to the root connector over a plurality of lanes comprising a first set of lanes and a second set of lanes, wherein the first set of lanes are associated with a first virtual communication channel and a second set of lanes are associated with a second virtual communication channel; a first accelerator communicatively coupled to the interconnect switch and associated with a first traffic class identifier corresponding to first communication traffic communicated over the first set of lanes; and a plurality of accelerators communicatively coupled to the interconnect switch and associated with a second traffic class identifier that corresponds to second communication traffic having lower priority than the first communication traffic and that is communicated over the second set of lanes.

Systems and methods for allocating bandwidth across a cluster of accelerators

Growing community of inventors

San Mateo, CA, United States of America

Lingling Jin