Lightning Talk: GPU.x - Dynamic GPU Fraction and Sharing on ML Mainstream Framework
summary
Claims a working PoC called GPU.x that injects a transparent Python hooker into mainstream ML frameworks to slice an 8 GB GPU into live fractions (demo shows ~30% and ~20% partitions). It intercepts framework-level allocation calls, enforces per-environment memory limits, lets partitions be reconfigured dynamically, and requires no code changes to models. Demo allocations succeed inside each partition and return OOM when a process exceeds its assigned fraction.
AI-written summary · about
related talks
09mLightning Talk: Bringing Multi-Tenancy to GPU-Based Workloads
09mDemocratizing Access to GPUs for Faster MLAI Innovation with Generative AI
09m⚡ Lightning Talk: Advanced GPU-Orchestrated Workflows and HPC Integrations on K8s for Distributed AI/ML at Scale
30mGPU Sharing at CERN: Cutting the Cake Without Losing a Slice!
15mTransparent GPU Sharing in Container Clouds for Deep Learning Workloads
10mLightning Talk: Profiling and Memory Debugging Tools for Distributed ML Workloads on GPUs
29mSharing is Caring – Fractional GPU Allocations With MetaGPU Device Plugin
09mLightning Talk: GPU-Scanner: Extending CNCF Observability for Multi-GPU AI Workloads