Platform skin temperature management in computer systems
Aspects of the disclosure are directed to a thermal management system for a computer system. In accordance with one aspect, the disclosure includes generating a transfer function representation of a relationship between a plurality of temperature data sequences and a plurality of dc power consumption data sequences; optimizing a dc power assignment to generate an optimized dc power assignment using the transfer function representation with minimization of a cost function; and distributing a plurality of dc power allocations to a plurality of components in a computer platform using the optimized dc power assignment.
1 . An apparatus comprising:
a plurality of thermal sensors configured to measure a plurality of temperature data sequences;
a plurality of dc power monitors configured to measure a plurality of dc power consumption data sequences; and
a system on a chip (SoC) configured to generate a z-transform transfer function representation of a relationship between the plurality of temperature data sequences and the plurality of dc power consumption data sequences.
2 . The apparatus of claim 1 , wherein the system on a chip (SoC) is further configured to optimize a dc power assignment using the z-transform transfer function representation.
3 . The apparatus of claim 2 , wherein the system on a chip (SoC) is further configured to optimize with minimization of a cost function.
4 . The apparatus of claim 3 , further comprising a computer platform, and wherein the system on a chip (SoC) is further configured to distribute a plurality of dc power allocations to a plurality of components in the computer platform using an optimized dc power assignment.
5 . The apparatus of claim 4 , wherein the system on a chip (SoC) is further configured to ingest the plurality of temperature data sequences from the plurality of thermal sensors.
6 . The apparatus of claim 4 , wherein the system on a chip (SoC) is further configured to ingest the plurality of dc power consumption data sequences from the plurality of dc power monitors.
7 . An apparatus for implementing dc power assignment optimization, the apparatus comprising:
means for ingesting a plurality of temperature data sequences from a plurality of thermal sensors in a computer platform;
means for ingesting a plurality of de power consumption data sequences from a plurality of dc power monitors in the computer platform;
means for generating a z-transform transfer function representation of a relationship between the plurality of temperature data sequences and the plurality of dc power consumption data sequences;
means for optimizing a dc power assignment using the z-transform transfer function representation with minimization of a cost function; and
means for distributing a plurality of dc power allocations to a plurality of components in the computer platform using the optimized dc power assignment.
8 . A method comprising:
generating a z-transform transfer function representation of a relationship between a plurality of temperature data sequences and a plurality of dc power consumption data sequences;
optimizing a dc power assignment to generate an optimized dc power assignment using the z-transform transfer function representation with minimization of a cost function; and
distributing a plurality of dc power allocations to a plurality of components in a computer platform using the optimized dc power assignment.
9 . The method of claim 8 , further comprising ingesting the plurality of temperature data sequences from a plurality of thermal sensors in the computer platform.
10 . The method of claim 9 , further comprising ingesting the plurality of dc power consumption data sequences from a plurality of dc power monitors in the computer platform.
11 . The method of claim 10 , wherein the z-transform transfer function representation includes a plurality of model parameters.
12 . The method of claim 11 , wherein the plurality of model parameters includes a pole of the z-transform transfer function representation.
13 . The method of claim 11 , wherein the plurality of model parameters includes a gain of the z-transform transfer function representation.
14 . The method of claim 11 , wherein the plurality of model parameters includes an ambient temperature.
15 . The method of claim 11 , wherein the cost function is subject to a plurality of maximum temperature constraints.
16 . The method of claim 15 , wherein the plurality of maximum temperature constraints includes one or more constraint parameters which are functions of the plurality of model parameters.
17 . The method of claim 8 , wherein the cost function is a weighted quadratic error of a dc power consumption relative to a target dc power allocation for the plurality of components using a plurality of optimization weights.
18 . The method of claim 17 , further comprising selecting the plurality of optimization weights to bias the optimization selectively among the plurality of components.
19 . The method of claim 17 , wherein the plurality of optimization weights are based on a dynamic workload (WL) of the plurality of components.
20 . The method of claim 8 , wherein the optimized dc power assignment is solved in polynomial time.