Modal Clusters goes GA: instant RDMA-connected GPU nodes billed by the second via one decorator
josh_wills · x · 2026-10-02
Modal has announced general availability of Modal Clusters, a new primitive two years in the making (1.5 years of battle-testing): users spin up globally available, RDMA-connected multi-node clusters instantly with a single @modal.clustered(size=4, rdma=True) decorator, billed by the second.
- The official example defines a training function on 8x B300 GPUs, with cluster.containerips() and containerrank() APIs exposing rank/world-size info for distributed training
- Clusters integrate with existing Modal primitives: checkpoints in Volumes, data via Cloud Bucket Mounts, orchestration via Queues
- Pitch: own your intelligence without owning the hardware, supporting petaFLOP-scale training and serving
Related event: Modal Clusters Reaches GA: Second-Billed RDMA GPU Clusters via a Decorator(2 posts)→
More from Infra
- PyTorch's TLX-based JFA kernel beats FlashAttention-4 by 13% fwd, 50% bwd on B200 — PyTorch · 2026-10-02
- Fireworks shows numerical mismatch can collapse RL training in 25 steps on GLM and MoE models — sophiamyang · 2026-10-02
- $13B Baseten bets on open models, launches Base Labs research lab — baseten · 2026-10-02
- Kaigen details Unity benchmark setup: Runtime Speed with native C/C++ multithreading — gdechichi · 2026-10-02
- a16z: every $100 into AI buildout sends $50 to chips, $20 to power; 100+ charts — demian_ai · 2026-10-02
- Workload-aware inference: why batch LLM pipelines should plan queries like databases do — sh_reya · 2026-10-02