dstack vs SkyPilot

dstack is an open-source orchestration layer for AI workloads across GPU clouds, Kubernetes, virtual machines, and bare-metal clusters. YAML configurations define fleets, development environments, tasks, services, presets, and volumes, while dstack handles provisioning and job scheduling, including auto-scaling, port forwarding, and ingress. Supported accelerators include NVIDIA, AMD, TPU, and Tenstorrent. Documented backends include AWS, Azure, GCP, Kubernetes, GPU cloud providers, remote SSH hosts, and experimental Slurm. Tasks can use frameworks such as accelerate, torchrun, Ray, and Spark. Users manage resources through the CLI or HTTP API, and the server can run on a laptop or another environment that can access the clusters in use. Services can publish inference endpoints through gateways with HTTPS, custom domains, auto-scaling, and rate limits. The self-hosted open-source stack is free. Server data and project secrets are stored in plaintext by default unless administrators configure AES-256-GCM encryption. TPU support is limited to single-host instances with up to eight cores. GPU Marketplace usage is prepaid, with prices that vary by GPU and provider and appear in the console before provisioning.

SkyPilot is an AI compute platform that manages Kubernetes, Slurm, virtual machines, on-premises systems and cloud resources through one interface. Listed workloads include development, pre-training, post-training and reinforcement learning, batch inference, sandboxes and inference endpoints. Scheduling provides priority queues, GPU sharing, quotas and intelligent placement across workloads. Fleet health checks can run proactively or reactively, with automatic remediation for GPU and NCCL faults. Integrations include AWS, GCP, Azure, Nebius, CoreWeave, Lambda, RunPod, Kubernetes, Slurm and on-premises infrastructure. SkyPilot can run standalone or connect to a remote API server for multi-user collaboration. Enterprise controls include SSO onboarding, RBAC, workspaces, policy enforcement, a unified dashboard, idle-compute stopping and cost reporting. Deployment options include BYOC, BYOK, private VPC and air-gapped environments. The project began as open source at UC Berkeley's Sky Computing Lab and is offered as open source. Documentation says Slurm support remains under active development and SkyServe on Slurm is not supported.