- Introduces Disaggregated Prefill and Decode (DPD) for SageMaker HyperPod, separating LLM prefill and decode onto dedicated GPU pools via EFA RDMA
- Provides consistent per-token latency, higher goodput, and independent scaling of prefill/decode workloads with an intelligent router
- Enables DPD via a `pdSpec` in the existing InferenceEndpointConfig, composable with KV cache offloading
Amazon SageMaker HyperPod now supports disaggregated prefill and decode
Official AWS changelog summary with source attribution, release tags, and community reactions.
Track this AWS release note alongside related changelog updates, risky changes, and weekly digest signals from the developer community.
Verify the original release note
Every summary should remain traceable to the original changelog or release source.
Follow this in context
Turn this AWS release note into a weekly digest for the tools you actually use.
Create digestMore AWS changelog updates
AWS Local Zone in Las Vegas, Nevada is now generally available, extending core services to the metropolitan area.
Adds support for AWS KMS customer‑managed keys to encrypt data at rest for InfluxDB 2, read replicas, and InfluxDB 3 clusters.
Amazon EC2 P6-B300 instances are now available in the Asia Pacific (Seoul) region.