RaptAI

GPUs are expensive, and when underutilized, it results in high total cost of ownership (TCO). rapt.ai software creates a virtual pool of all GPUs available that enables HPE Ezmeral Runtime users to run multiple training jobs in parallel based on dynamic resource demands and business workflows.

raptIQ is rapt.ai’s compute automation platform designed to deliver improved performance and utilization for GPU-dependent workloads. raptIQ supports a broad range of popular GPU devices and can abstract these resources dynamically based on workload needs. raptIQ’s adaptive scheduler can learn and intelligently pack, auto pre-empt and schedule AI models based on user priorities and business workflows. Data scientists and developers can leverage the combination of HPE Ezmeral Runtime and raptIQ to deliver faster time to accuracy for the AI/ML models and with better GPU resource usage.

As demand for AI/ML workloads increase, providing efficient and granular GPU access to teams of researchers running hundreds of jobs on many clusters over weeks is quickly becoming a huge challenge. rapt.ai on HPE Ezmeral Runtime addresses this challenge and enables:

• Improve GPU resource automation and allocation– supports a broad range of popular GPU devices and can abstract these resources dynamically based on workload needs
• Achieve faster model accuracy - automating the scheduling of workloads and creating policies and priorities that dynamically allocate resources
• Flexible GPU allocation methods – supports auto or manual mode to allocate GPU resources.  Auto-mode will auto-analyze workloads and dynamically allocate required GPU resources
• Workload visibility – Dashboard displays visibility into GPU utilization, model history, and other key GPU usage statics to enable deep insights of the running workloads