Thermal Intelligence
for Data Centers
& Industry.
Everyone else measures the room. We calculate the chip.
arflo connects thermal physics and adaptive AI to understand operating conditions, from the chip to the infrastructure.
- Selected rack
- R02
- GPU temperature
- 62.0 °C
- Control mode
- Adaptive
Less cooling energy
Understand cooling demand in the context of load and thermal conditions.
Predictive Maintenance
Investigate changing thermal behavior and assess emerging maintenance needs.
Full Transparency
Understand signals, model assumptions and control decisions.
Your perspective.
Your starting point.
Operations, product integration, or energy and heat use: explore the questions and applications relevant to you.
Data Center Operators & Customers
OPERATIONS / COLOCATION / COMPUTEConnect compute demand, thermal behavior and cooling decisions across your operating context.
02OEMs
EQUIPMENT / PRODUCT INTEGRATIONExplore how thermal intelligence could support the equipment you design, build and integrate.
03Energy & Heat
HEAT REUSE / ENERGY / GRIDExplore thermal and load profiles for heat-reuse planning, energy supply and grid-related questions.
Intelligence follows heat.
It starts with compute.
Every workload changes the power demand of your infrastructure. And the heat your cooling system needs to remove.
Same workload.
Two control strategies.
What happens when cooling follows compute? Explore three load profiles in a simplified thermal model and compare temperature and cooling electricity demand.
- IT load
- 95 %
- Static
- 61.3 °C
- Adaptive
- 60.1 °C
View chart data as a table
| Time | IT load | Static | Adaptive |
|---|---|---|---|
| 0 min | 38% | 60.0 °C | 60.0 °C |
| 5 min | 38% | 46.3 °C | 58.9 °C |
| 10 min | 92% | 48.7 °C | 59.3 °C |
| 15 min | 95% | 61.3 °C | 60.1 °C |
| 20 min | 41% | 58.8 °C | 60.1 °C |
| 25 min | 38% | 46.1 °C | 58.9 °C |
| 30 min | 38% | 43.3 °C | 58.4 °C |
Solid line: adaptive control. Dashed line: static operation. Scroll the chart horizontally when needed, or use the data table.
Simplified single-node model, not measured data or a savings forecast. Same initial temperature (60 °C), same inlet (22 °C), identical load. Model parameters are illustrative; actual results depend on the system.
What makes a fair comparison ↗︎One model.
The whole system.
The engineering idea behind arflo: connect operational data with thermal understanding and derive appropriate cooling control inputs.
Your infrastructure
Temperatures · Compute load · Cooling state
arflo
Physical model + adaptive optimization
Your cooling
Setpoints · Control inputs · Feedback
Understand the data.
Which signals are available? We establish measurement points, data quality and your system interfaces.
Evaluate the model.
Predictions and measurements are compared. Load changes and model limitations belong in the evaluation.
Plan control together.
Control inputs, approvals and fallback behavior are defined with the operations team for the specific system.
Integration scope, model deployment and access to controls are agreed for each pilot project.
Discuss integration ↗︎Research &
publications.
Four technical papers on thermal models, sensing and control. With explicit derivations, worked examples and references, available online and as PDFs.
Thermal networks: why the chip has more than one time constant.
A coupled RC model explains how a load step travels through a package, and why local resistance-capacitance products do not directly describe the complete transient.
Read paper · 8 MIN · PDFSmaller thermal models that preserve the physics.
Projection can preserve thermal dissipation and reduce computational size. A worked counterexample shows why stability alone does not guarantee accurate chip temperatures.
Read paper · 8 MIN · PDFThermal estimation on the measurement clock.
A timestamp-aware formulation for combining power and temperature signals with different sampling intervals and transport delays.
Read paper · 10 MIN · PDFA physics-based filter for cooling commands.
An exact one-step thermal constraint that separates a controller’s preferred action from the heat removal needed to respect a stated temperature limit.
Read paper · 10 MIN · PDFEngineering notes
Chip telemetry as an input for cooling control.
Servers already report temperatures, power draw and utilization for their own protection. This note looks at what that telemetry contains, how it can inform facility cooling and what has to be true before a controller relies on it.
Read note · 4 MINSetpoint control in direct-to-chip liquid cooling.
A direct-to-chip liquid cooling loop has two main control inputs: coolant supply temperature and flow. This note explains what each one does, which limits apply and why the best setting changes with load.
Read note · 5 MINGPU load transients and cooling control.
GPU clusters change their heat output faster than most cooling loops can follow. This note explains where the delay comes from and how feedforward control uses the load signal to close the gap.
Read note · 5 MINHow AI optimizes data center cooling.
AI data center cooling optimization means adjusting cooling setpoints continuously to the actual heat load. This note explains what gets optimized, which methods exist and what to check before one of them touches a live system.
Read note · 6 MINGet more
from your
infrastructure.
Show us your challenge. We’ll show you how arflo can make your cooling smarter.