What is Foundry Cloud Platform?
Foundry Cloud Platform provides burst compute capabilities for machine learning teams. It allows users to quickly access NVIDIA GPUs for training, fine-tuning, and inference tasks without long-term contracts or unreliable hardware.
The platform supports both spot and reserved instances. Spot instances offer significant cost savings for flexible workloads, while reserved instances provide guaranteed performance and availability. Users can manage instances programmatically and benefit from features such as custom scripts, disk state saving, and integration with Kubernetes.
Features
- Spot Instances: Cost-efficient compute for flexible training, fine-tuning, and inference.
- Reserved Instances: Guaranteed performance and reliability for training, fine-tuning, and inference.
- Programmatic Scaling: Place spot orders via API to automate scaling.
- Kubernetes Integration: Simplify workload orchestration and horizontal scaling with Kubernetes.
- NVIDIA Tensor Core GPUs: Access NVIDIA H100s, A100s, A40s, and A5000s without a contract.
- High-Performance Networking: Optimize distributed training with 1.6 Tbps Infiniband.
- Co-located Storage: Access storage with no ingress/egress fees.
Use Cases
- Machine learning model training
- Fine-tuning pre-trained models
- Running inference workloads
- Scaling compute resources for AI research
- Managing AI infrastructure costs
