High-Throughput CALPHAD Screening (HTCS) generates massive, multi-dimensional alloy datasets that exceed the rendering and memory capacities of CPU (Central Processing Unit) and Document Object Model (DOM) based visualization tools. When processing millions of alloy data simulations, existing web technologies will, at times, experience network payload failures, memory exhaustion, and visual overplotting. This study implements a GPU (Graphics Processing Unit) accelerated visual analytics architecture to address these limitations.

The system bypasses DOM manipulation by packaging all data into a 72 MB flat binary buffer, which is transferred directly to the GPU’s Video Random Access Memory (VRAM). WebGL (Web Graphics Library) is used to transform a discrete Parallel Coordinates Plot (PCP) into a Density-Based Parallel Coordinates (DBPC). This is achieved using a 32-bit floating-point accumulation buffer and additive blending, which converts overlapping data paths into continuous multi-dimensional heatmaps.

The results of the system evaluation benchmarks show that the proposed GPU architecture processes 3 million data points with a stable render time of approximately 743 milliseconds without any browser crash. The visual analysis objectively demonstrates that the DBPC model reduced maximum pixel saturation from 66.16% to 7.33%. This reduction resulted in a 340% increase in preserved structural information compared to the baseline 8-bit alpha blending, with a corresponding increase in Shannon Entropy from 1.27 bits/px to 5.59 bits/px. A practical case study in automotive steel confirmed the tool functionality, showcasing the ability of interactive, real-time vertex culling. Domain experts can use real-time filtering to reliably isolate viable alloy candidates and identify stable thermodynamic properties and trade-offs.