They are used for doom scrolling and social media.

They have the computational power of a laptop, and sit unused. That is a waste.

What is stopping us? Heat. I aim to solve that.

I have gotten my code's thermal prediction accuracy to a remarkable level, and was able to launch and sustain an animation rendering Discord bot with real time physics simulations and heavy cache operations and computational backend. My launcher successfully deferred operations before reaching throttle temperature, predicted thermal events before they happened, and during a stress test where I launched my bot quickly to overheat my phone, my launcher shut down my bot before it reached danger level temperature.

UPDATE (Nov 5, 2025):

Performance Numbers (1 hour production test on Discord bot serving 645+ members):

============================================================ PREDICTION ACCURACY ============================================================ Total predictions: 21372 MAE: 1.82°C RMSE: 3.41°C Bias: -0.38°C Within ±1°C: 57.0% Within ±2°C: 74.6%

Per-zone MAE: BATTERY : 1.68°C (3562 predictions) CHASSIS : 1.77°C (3562 predictions) CPU_BIG : 1.82°C (3562 predictions) CPU_LITTLE : 2.11°C (3562 predictions) GPU : 1.82°C (3562 predictions) MODEM : 1.71°C (3562 predictions) ============================================================

I don't know about everyone else, but I didn't want to pay for a server, and didn't want to host one on my computer. I have a flagship phone; an S25+ with Snapdragon 8 and 12 GB RAM. It's ridiculous. I wanted to run intense computational coding on my phone, and didn't have a solution to keep my phone from overheating. So. I built one. This is non-rooted using sys-reads and Termux (found on Google Play) and Termux API (found on F-Droid), so you can keep your warranty.

Just for ease, the repo is also posted up here.

https://github.com/DaSettingsPNGN/S25_THERMAL-

What my project does: Monitors core temperatures using sys reads and Termux API. It models thermal activity using Newton's Law of Cooling to predict thermal events before they happen and prevent Samsung's aggressive performance throttling at 42° C.

Target audience: Developers who want to run an intensive server on an S25+ without rooting or melting their phone.

Comparison: I haven't seen other predictive thermal modeling used on a phone before. The hardware is concrete and physics can be very good at modeling phone behavior in relation to workload patterns. Samsung itself uses a reactive and throttling system rather than predicting thermal events. Heat is continuous and temperature isn't an isolated event.

I didn't want to pay for a server, and I was also interested in the idea of mobile computing. As my workload increased, I noticed my phone would have temperature problems and performance would degrade quickly. I studied physics and realized that the cores in my phone and the hardware components were perfect candidates for modeling with physics. By using a "thermal bank" where you know how much heat is going to be generated by various workloads through machine learning, you can predict thermal events before they happen and defer operations so that the 42° C thermal throttle limit is never reached. At this limit, Samsung aggressively throttles performance by about 50%, which can cause performance problems, which can generate more heat, and the spiral can get out of hand quickly.

My solution is simple: never reach 42° C

https://github.com/DaSettingsPNGN/S25_THERMAL-

Please take a look and give me feedback.

Thank you!