US 12,468,616 B2
Generating power efficient secure multiclient performance monitoring data
Pranav Vaidya, San Jose, CA (US); Alan Menezes, San Jose, CA (US); Siddharth Sharma, San Jose, CA (US); Jin Ouyang, Cupertino, CA (US); Gregory Paul Smith, Leander, TX (US); Timothy J. McDonald, Austin, TX (US); Shounak Kamalapurkar, Santa Clara, CA (US); Abhijat Ranade, Austin, TX (US); and Thomas Melvin Ogletree, Lakeway, TX (US)
Assigned to NVIDIA CORPORATION, Santa Clara, CA (US)
Filed by NVIDIA CORPORATION, Santa Clara, CA (US)
Filed on Mar. 18, 2022, as Appl. No. 17/698,668.
Prior Publication US 2023/0297485 A1, Sep. 21, 2023
Int. Cl. G06F 11/34 (2006.01); G06F 1/10 (2006.01); G06F 21/60 (2013.01)
CPC G06F 11/3409 (2013.01) [G06F 1/10 (2013.01); G06F 21/602 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method for generating performance monitoring data, the method comprising:
receiving an electronic signal representing a performance metric of a processor included in a subset of processors included in a parallel processing system;
incrementing a first unit level counter associated with the performance metric during each clock cycle in which the electronic signal is at a first state;
serializing a value stored in the first unit level counter to generate a serialized value; and
transmitting the serialized value via one or more electronic signal paths,
wherein a collection unit comprising a set of counters:
generates an aggregated counter value of the serialized value stored in the first unit level counter and other counter values,
stores the aggregated counter value into an active memory bank, and
upon receiving a trigger condition, switches the active memory bank with an idle memory bank, such that the aggregated counter value is now stored in the idle memory bank.