RunPod

RunPod is an all-in-one cloud for developing, training and running AI models. The service offers a globally distributed GPU cloud that lets customers run any GPU workload and focus on machine learning models, not the underlying infrastructure.

With RunPod, customers can spin up GPU pods in seconds with a variety of GPU options, including MI300X, H100 PCIe, A100 PCIe and more. Each GPU instance is billed by the minute, with no ingress or egress fees. The service also supports serverless ML inference, which means autoscaling and job queueing with sub-250ms cold start times.

RunPod also offers instant hot-reloading for local changes, so customers can test and deploy without interruption. Customers can use over 50 preconfigured templates for popular frameworks like PyTorch and Tensorflow or bring their own custom containers. Provisioning and deployment is done through a CLI tool.

Some of the key features of RunPod include:

• GPU Cloud: Spin up GPU pods in seconds with a wide range of options.
• Serverless ML Inference: Autoscale and job queueing with sub-250ms cold start times.
• Instant Hot-Reloading: Test and deploy changes seamlessly.
• Preconfigured Templates: Over 50 templates for popular frameworks like PyTorch and Tensorflow.
• Custom Containers: Bring your own custom containers for deployment.
• CLI Tool: Streamlined provisioning and deployment.
• 99.99% Uptime: High availability for demanding workloads.
• 10PB+ Network Storage: Scalable storage options.
• Real-Time Logs and Analytics: Detailed metrics for debugging and optimization.

RunPod pricing is based on the type of GPU instance and usage. For example, the MI300X GPU instance costs $4.89 per hour, while the RTX A4000 Ada costs $0.39 per hour. Storage is billed at $0.10 per GB per month for running pods and $0.20 per GB per month for idle pods.

RunPod has secured $20M in funding to further accelerate the transformation of AI/ML cloud computing. It's designed to be the developer's launchpad for full-stack AI applications, with the goal of becoming the compute backbone that allows successful companies to run AI/ML workloads at scale.

Published on June 14, 2024