Why Thermistor

As users of mobile workstations we believe that instead of controlling performance, people expect a control over:

• Surface temperature of the device (with distinction between bottom & keyboard)
• Fan loudness
• Battery usage

and operate at maximum performance with those constraints.

The project is mainly focused on mobile workstations/gaming laptops like Dell XPS, Dell Precision, Lenovo Legion, Asus ROG etc., but is also applicable to any other laptop and tower device.

Switching Context

During an office day, a typical user of a mobile workstation may switch the way of operating their device between contexts, which implies different preferences.

Docked The device operates as a tower computer, a user may only care about Fan loudness to not disturb others (especially in case of very loud fan). The lid is usually closed.

Powered The device is connected to mains power and used directly. User cares about:

1. Keyboard surface temperature
2. Fan loudness as above

Meeting The device lies on the table and is battery powered. The user now cares about Battery usage as the main point.

Lap-top The device lies on lap, usually battery powered. Priorities:

1. Bottom surface temperature
2. Top surface temperature
3. Battery usage
4. Fan loudness

The current state

Mobile workstations vary a lot in their cooling performance. Nevertheless, producers usually pack and advertise strong CPU and GPU, which cannot be sustainably cooled at their TDP. This often leads to overthrottling in such devices. For example, Dell XPS 15 9560 from 2017 has Intel i7-7700HQ (45W TDP) and NVIDIA GeForce GTX 1050 Mobile (~50-70W TDP). There is not way for a laptop with so small heat sink to dissipate 100W of heat. On the other hand, in order to support lap-top usage, the bottom notebook surface is partially thermically isolated. That forces the built-in PIDs controllers to be very conservative.

Goal of the project

Thermistor project aims to look at the aforementioned issue from heat dissipation perspective with specific goals to:

• measure,
• predict,
• detect,
• apply,
• help improve.

Thermistor should care about power budget, and boosting, throttling, frequencies should be managed by existing embedded controllers.

TODO

• Research reliably setting L1/L2 power limits + measuring package power in Intel CPUs (is intel-rapl enough?)
• Find reliable API to collect and classify current status:
  • Temperature: knowing temperature at various points in the device: CPU/GPU package, disk, battery, peripherals.
  • Estimate environment temperature (location? sensors while idle?)
  • Fan speed
  • Fan loudness (microphone? notebookcheck reviews?)
  • CPU state: freqs, active boost, c-states, power consumption
  • GPU state: same
  • Embedded Controller throttling reason
  • Total power draw: from battery, from charger, charger caps (e.g. minimal 65W USB-PD or stock 160W)
• Research controllability of mobile Nvidia GPUs power limits + measuring current draw.
• Find API for controlling the fans if possible.
• Build a simple feedback thermometers for laboratory work.
• Learn to measure & model heat dissipation + heat capacity in various settings.
  • Remove the external sensors from the model.
• Apply the model to control power limits dynamically.
• Detect anomalies to warn users when heat dissipation is far from expected.
• Check the predictability of power-budgeting the battery.
• Investigate the ways of increasing heat dissipation capacity:
  • Make a benchmark for comparing cooling pads.
  • Test the effect of increasing thermal conductance to the bottom surface: vanilla, with external cooling pad, on lap.
  • Try to build detachable external heat sink for docked setup.
  • Build a gadget that recycles the heat.

Awesome list of possible backends

• https://github.com/kitsunyan/intel-undervolt C with support for RAPL, configured through /etc file

• https://github.com/georgewhewell/undervolt Python, understands ThrottleStop

• Also, check out the wiki if interested in reading random research-dumps.