Publication Title: Thermo-Algorithmic Determinism in HPC Clusters: Mitigating the Straggler Effect through Proactive Hydrodynamic Telemetry Acronym: TACE (Thermo-Algorithmic Control & Execution)

Publication Metadata (Zenodo) Title: Thermo-Algorithmic Determinism in Hyperscale AI Architectures: The mCDU Data Physics Framework Project Acronym: TACE (Thermo-Algorithmic Control & Execution) DOI: [https://doi.org/10.5281/zenodo.22997888](https://doi.org/10.5281/zenodo.22997888) Keywords: High-Performance Computing (HPC), Artificial Intelligence, Liquid Cooling (mCDU), Cyber-Physical Systems, Thermo-Algorithmic Determinism, DVFS, Straggler Effect, Fluid Dynamics, Green Computing. Publication Description (Abstract & Repository Overview) The scalability and performance of multi-node AI accelerator clusters (e.g., NVIDIA H100/GB200 architectures) are critically dependent on microsecond-level synchronization across distributed environments (e.g., NCCL Ring-AllReduce). This publication identifies and formalizes a physical-layer vulnerability—a "Hardware Zero-Day"—inherent in modern liquid cooling infrastructure (mCDU). The research demonstrates how standard, reactive industrial automation (BMS/PLC) inadvertently exacerbates static suction pressure drops during computational load spikes. This leads to cavitation erosion, gas desorption governed by Henry's Law, and the physical blockage of processor microchannels, a phenomenon known as bubble pinning. The resulting cascading hardware clock throttling (DVFS) introduces millisecond-scale latencies that paralyze the throughput of the entire cluster, an issue recognized as the Straggler Effect. In response to this challenge, this work defines the framework of Thermo-Algorithmic Determinism and introduces the TACE algorithm, serving as the first fully software-based patch for hardware hydraulic vulnerabilities. This document presents the complete mathematical evolution of the phenomenon, deriving a rigorous, coupled 0-Dimensional Lumped-Parameter Differential-Algebraic Equation system. For the first time in the literature, this model bridges nanosecond-scale instruction execution logic with the millisecond-scale hydrodynamic inertia of coolant phase transitions. Core Theoretical Frameworks Documented in the Repository: Deterministic Stopping Condition: Transforms the algorithm execution time from an implicit cycle formulation into an explicit function conditioned by variable RAM latency during frequency throttling. Coupled State System: Strictly links computational kinematics, silicon thermal balance (enriched with smooth idle-state transition functions and temperature-driven transistor leakage currents), and delayed boiling kinetics. Proactive Hydrodynamic Telemetry (TACE Core Logic): A control mechanism based on evaluating the time derivative of differential pressure variance. By analyzing non-linear collapses in the fluid's bulk modulus within a high-frequency telemetry window, the algorithm pre-emptively decelerates pump drives. This restores static suction pressure (NPSHa) well before microchannel colmatation can occur. Green Computing Optimization: An integrated, macroscopic facility energy cost model that utilizes two-phase friction multipliers for fluid mixtures to accurately estimate total pump workload during hydrodynamic breakdowns.