Research Dossier / 01

Parallel Clique Optimization on GPUs

Research focused on accelerating maximal clique finding in large-scale graphs through novel GPU-based parallelization strategies, significantly optimizing computational complexity for dense graph structures.

Short LabelGPU Clique
Institution
Research under Dr. Vishwesh Jatala, IIT Bhilai
Role
Researcher
Active Period
2023 - Present
CUDA
GPU Programming
Parallel Computing
Graph Algorithms

Method + iterations

Research steps, prototypes, and refinements along the way.

  1. Redesigned traditional clique optimization algorithms for massive parallel execution on CUDA

    enabled GPUs.

  2. Introduced optimized thread

    block mapping strategies to reduce warp divergence and memory bottlenecks.

  3. Implemented shared

    memory utilization techniques for efficient neighborhood intersection computation.

  4. Built a graph preprocessing stage for node re

    ordering and compact CSR — style adjacency representation.

  5. Added adaptive kernel launch policies based on graph density profile and frontier size.

  6. Benchmarked performance improvements against CPU

    based and baseline GPU implementations across synthetic and real — world datasets.

Visual Notes

GPU PARALLELIZATION

GPU PARALLELIZATION

Optimized thread scheduling and memory coalescing for large nodal graph clique minimization.

KERNEL PIPELINE

KERNEL PIPELINE

Multi-stage kernel pipeline with preprocessing, candidate filtering, and iterative clique expansion.

PERFORMANCE ANALYSIS

PERFORMANCE ANALYSIS

Comparative profiling against CPU and baseline GPU implementations under varying graph densities.

← Back to archiveNext / GPU Triangles