---
title: "Scalable Multi-GPU Simulation of 3D Multicellular Growth with RNN-Based Workload Balancing"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2608.25890v1"
permalink: "https://feed.astralor.com/items/itm_6bca7a7c1fedc4093925925a"
published: "2026-08-26T15:06:10.000Z"
score: 70
channel: "ai"
tags: ["GPU Acceleration","RNN","Subcellular Element Model","Multicellular Simulation","Load Balancing"]
---

# Scalable Multi-GPU Simulation of 3D Multicellular Growth with RNN-Based Workload Balancing

## AI commentary

> What intrigues me about this paper is using an RNN as the controller for load balancing rather than purely predicting workload, offering a fresh perspective on dynamic scheduling. For engineers working on HPC and scientific simulation, especially GPU cluster scheduling, it is worth reading. That said, the validation scenario is somewhat biased toward biological simulation, so broader cross-domain generalization remains to be demonstrated.

## At a glance

Detailed 3D multicellular growth simulations based on subcellular element models (SEMs) incur substantial computational costs. This paper presents a scalable multi-GPU framework that combines GPU acceleration, spatial binning, domain decomposition, and workload-aware partitioning. Because cell movement, growth, and division continuously reshape the spatial workload distribution, initially balanced partitions become inefficient over time. To address this, the authors introduce an RNN-based load-balancing controller that observes recent per-rank execution times and partition states, and learns residual corrections to a reactive boundary-adjustment rule. The controller is trained offline in a differentiable surrogate of the load-balancing loop with randomized workload dynamics, requiring no measured execution traces. Evaluation shows that GPU acceleration with spatial binning boosts interaction computation by roughly three orders of magnitude over a serial CPU baseline. RNN-guided load balancing reduces mean global imbalance from 11.3% under static partitioning to 3.5%, lowers end-to-end runtime by 9.0% relative to static partitioning, and reduces slice migration by 7.7x compared with the reactive baseline, demonstrating that history-aware control improves workload balance while avoiding unnecessary repartitioning.

## Key points

- A scalable multi-GPU framework for 3D multicellular growth simulation is proposed, integrating GPU acceleration, spatial binning, domain decomposition, and workload-aware partitioning.
- An RNN-based load-balancing controller is introduced, trained offline to learn residual corrections beyond reactive rules without requiring measured execution traces.
- GPU acceleration with spatial binning speeds up interaction computation by roughly three orders of magnitude over a serial CPU baseline.
- RNN-guided load balancing reduces global imbalance from 11.3% to 3.5%, lowers end-to-end runtime by 9.0%, and cuts slice migration by 7.7x compared with the reactive baseline.

## Full text

Detailed multicellular growth simulations based on subcellular element models (SEMs) can capture complex tissue development, but their element-level interactions impose substantial computational cost. This work presents a scalable multi-GPU framework for 3D multicellular growth simulation that combines GPU acceleration, spatial binning, domain decomposition, and workload-aware partitioning. Cell movement, growth, and division continuously reshape the spatial workload distribution, causing initially balanced partitions to become inefficient over time. To address this, we introduce an RNN-based load-balancing controller that observes recent per-rank execution times and partition states and learns residual corrections to a reactive boundary-adjustment rule. The controller is trained offline in a differentiable surrogate of the load-balancing loop with randomized workload dynamics, requiring no measured execution traces for training. We evaluate the framework in terms of single-GPU acceleration, multi-GPU computation scaling, controller-level load-balancing behavior, and end-to-end simulation performance, with comparisons against static partitioning, reactive load balancing, and conventional time-series prediction baselines. A representative embryonic epidermal development use case further demonstrates the type of spatially and temporally evolving workload targeted by the framework. In our evaluation, GPU acceleration with spatial binning accelerates the interaction computation by roughly three orders of magnitude over a serial CPU baseline. RNN-guided load balancing reduces the mean global imbalance from 11.3% under static partitioning to 3.5%, lowers end-to-end runtime by 9.0% relative to static partitioning, and reduces slice migration by 7.7x compared with the reactive baseline, showing that history-aware control can improve workload balance while avoiding unnecessary repartitioning.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_6bca7a7c1fedc4093925925a