PalGPU: A Cost-Efficient Distributed GPU Computing Platform for AI Workloads in Palestine

Description

PalGPU is a proposed shared GPU computing platform that aims to improve access to affordable AI computing resources in Palestine. The platform will connect available GPU-enabled computers contributed by individuals, universities, laboratories, and technology companies, allowing users to submit supported AI and deep learning workloads through a centralized web application. The system will include GPU node registration, job submission and queues, resource-aware and cost-aware scheduling, real-time GPU and VRAM monitoring, cost estimation, execution history, budget alerts, and provider earnings tracking. A companion mobile application will enable users to monitor their jobs and receive notifications remotely. The project will also investigate multi-GPU training using existing technologies such as PyTorch Distributed Data Parallel, subject to hardware and network suitability. The project will evaluate the technical and economic feasibility of this resource-sharing model in Palestine. Evaluation will consider execution time, waiting time, throughput, GPU utilization, electricity consumption, and cost per completed job, alongside a comparison with selected international cloud GPU services and an assessment of potential local demand. The platform will also provide students with access to multi-CPU computing resources at the university’s Centre of Excellence, enabling supported computational workloads to run across multiple CPUs.