[Docs] response coordination between thermal monitoring

[seanmcilroy29]

Outline Action Item Details

From NTTData:

Regarding the table 1, response coordination between thermal monitoring and the cooling system, it will be difficult to realize the linear response coordination from the cooling equipment side, depending on the cooling methodology. It is necessary to consider how the cooling equipment responds and follows the server-side conditions.

Issue dependency with other TF Groups

No response

[jabariwg]

This is true! The physics of the cooling system fundamentally prevents the equipment from responding linearly to dynamic server-side conditions. This has always been the case though CPUs have historically had much lower average and peak power demands, so heat flux could be managed with good fan controls. GPUs, however, will require new approaches to thermal management, like liquid cooling, especially as TDP continue to rise.
The WDPC workgroup is proposing a solution that goes beyond the hierarchal demand-response framework for workload-power-cooling management. The "data-bus" could provide real-time state information, along with predictions of future states of workloads (for example). All devices (servers, thermal monitors, power infrastructure, cooling systems, etc) can then independently modify their individual operations in preparation for several future operational states.

[jabariwg]

Response provided. From jabariwg: This is true! The physics of the cooling system fundamentally prevents the equipment from responding linearly to dynamic server-side conditions. This has always been the case though CPUs have historically had much lower average and peak power demands, so heat flux could be managed with good fan controls. GPUs, however, will require new approaches to thermal management, like liquid cooling, especially as TDP continue to rise. The WDPC workgroup is proposing a solution that goes beyond the hierarchal demand-response framework for workload-power-cooling management. The "data-bus" could provide real-time state information, along with predictions of future states of workloads (for example). All devices (servers, thermal monitors, power infrastructure, cooling systems, etc) can then independently modify their individual operations in preparation for several future operational states. Le ven. 5 sept. 2025 à 05:03, Sean Mcilroy ***@***.***> a écrit :

…

Assigned #14 <#14> to @jabariwg <https://github.com/jabariwg>. — Reply to this email directly, view it on GitHub <#14 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/A64EBDEB7ESNBAEKTPF7PZL3RFGW5AVCNFSM6AAAAACFWQOHGKVHI2DSMVQWIX3LMV45UABCJFZXG5LFIV3GK3TUJZXXI2LGNFRWC5DJN5XDWMJZGUZDIOJSGI2DMNQ> . You are receiving this because you were assigned.Message ID: <Green-Software-Foundation/Workload-Dynamic-Power-and-Cooling/issue/14/issue_event/19524922466 @github.com>