{"id":439,"date":"2026-02-20T14:51:30","date_gmt":"2026-02-20T06:51:30","guid":{"rendered":"https:\/\/connectword.dpdns.org\/?p=439"},"modified":"2026-02-20T14:51:30","modified_gmt":"2026-02-20T06:51:30","slug":"nvidia-releases-dynamo-v0-9-0-a-massive-infrastructure-overhaul-featuring-flashindexer-multi-modal-support-and-removed-nats-and-etcd","status":"publish","type":"post","link":"https:\/\/connectword.dpdns.org\/?p=439","title":{"rendered":"NVIDIA Releases Dynamo v0.9.0: A Massive Infrastructure Overhaul Featuring FlashIndexer, Multi-Modal Support, and Removed NATS and ETCD"},"content":{"rendered":"

NVIDIA has just released Dynamo v0.9.0<\/strong>. This is the most significant infrastructure upgrade for the distributed inference framework to date. This update simplifies how large-scale models are deployed and managed. The release focuses on removing heavy dependencies and improving how GPUs handle multi-modal data.<\/p>\n

The Great Simplification: Removing NATS and etcd<\/strong><\/h3>\n

The biggest change in v0.9.0 is the removal of NATS<\/strong> and ETCD<\/strong>. In previous versions, these tools handled service discovery and messaging. However, they added \u2018operational tax\u2019 by requiring developers to manage extra clusters.<\/p>\n

NVIDIA replaced these with a new Event Plane<\/strong> and a Discovery Plane<\/strong>. The system now uses ZMQ (ZeroMQ)<\/strong> for high-performance transport and MessagePack<\/strong> for data serialization. For teams using Kubernetes, Dynamo now supports Kubernetes-native service discovery<\/strong>. This change makes the infrastructure leaner and easier to maintain in production environments.<\/p>\n

Multi-Modal Support and the E\/P\/D Split<\/strong><\/h3>\n

Dynamo v0.9.0 expands multi-modal support across 3 main backends: vLLM<\/strong>, SGLang<\/strong>, and TensorRT-LLM<\/strong>. This allows models to process text, images, and video more efficiently.<\/p>\n

A key feature in this update is the E\/P\/D (Encode\/Prefill\/Decode) split<\/strong>. In standard setups, a single GPU often handles all 3 stages. This can cause bottlenecks during heavy video or image processing. v0.9.0 introduces Encoder Disaggregation<\/strong>. You can now run the Encoder<\/strong> on a separate set of GPUs from the Prefill<\/strong> and Decode<\/strong> workers. This allows you to scale your hardware based on the specific needs of your model.<\/p>\n

Sneak Preview: FlashIndexer<\/strong><\/h3>\n

This release includes a sneak preview of FlashIndexer<\/strong>. This component is designed to solve latency issues in distributed KV cache<\/strong> management.<\/p>\n

When working with large context windows, moving Key-Value (KV) data between GPUs is a slow process. FlashIndexer improves how the system indexes and retrieves these cached tokens. This results in a lower Time to First Token (TTFT)<\/strong>. While still a preview, it represents a major step toward making distributed inference feel as fast as local inference.<\/p>\n

Smart Routing and Load Estimation<\/strong><\/h3>\n

Managing traffic across 100s of GPUs is difficult. Dynamo v0.9.0 introduces a smarter Planner<\/strong> that uses predictive load estimation<\/strong>.<\/p>\n

The system uses a Kalman filter<\/strong> to predict the future load of a request based on past performance. It also supports routing hints<\/strong> from the Kubernetes Gateway API Inference Extension (GAIE)<\/strong>. This allows the network layer to communicate directly with the inference engine. If a specific GPU group is overloaded, the system can route new requests to idle workers with higher precision.<\/p>\n

The Technical Stack at a Glance<\/strong><\/h3>\n

The v0.9.0 release updates several core components to their latest stable versions. Here is the breakdown of the supported backends and libraries:<\/p>\n

\n\n\n\n\n\n\n\n\n\n
Component<\/strong><\/td>\nVersion<\/strong><\/td>\n<\/tr>\n<\/thead>\n
vLLM<\/strong><\/td>\nv0.14.1<\/td>\n<\/tr>\n
SGLang<\/strong><\/td>\nv0.5.8<\/td>\n<\/tr>\n
TensorRT-LLM<\/strong><\/td>\nv1.3.0rc1<\/td>\n<\/tr>\n
NIXL<\/strong><\/td>\nv0.9.0<\/td>\n<\/tr>\n
Rust Core<\/strong><\/td>\ndynamo-tokens crate<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/figure>\n

The inclusion of the dynamo-tokens<\/strong> crate, written in Rust<\/strong>, ensures that token handling remains high-speed. For data transfer between GPUs, Dynamo continues to leverage NIXL (NVIDIA Inference Transfer Library)<\/strong> for RDMA-based<\/strong> communication.<\/p>\n

Key Takeaways<\/strong><\/h3>\n
    \n
  1. Infrastructure Decoupling (Goodbye NATS and ETCD)<\/strong>: The release completes the modernization of the communication architecture. By replacing NATS and ETCD with a new Event Plane<\/strong> (using ZMQ<\/strong> and MessagePack<\/strong>) and Kubernetes-native service discovery<\/strong>, the system removes the \u2018operational tax\u2019 of managing external clusters.<\/li>\n
  2. Full Multi-Modal Disaggregation (E\/P\/D Split)<\/strong>: Dynamo now supports a complete Encode\/Prefill\/Decode (E\/P\/D)<\/strong> split across all 3 backends (vLLM, SGLang, and TRT-LLM). This allows you to run vision or video encoders on separate GPUs, preventing compute-heavy encoding tasks from bottlenecking the text generation process.<\/li>\n
  3. FlashIndexer Preview for Lower Latency<\/strong> :The \u2018sneak preview\u2019 of FlashIndexer<\/strong> introduces a specialized component to optimize distributed KV cache<\/strong> management. It is designed to make the indexing and retrieval of conversation \u2018memory\u2019 significantly faster, aimed at further reducing the Time to First Token (TTFT).<\/li>\n
  4. Smarter Scheduling with Kalman Filters<\/strong>: The system now uses predictive load estimation<\/strong> powered by Kalman filters<\/strong>. This allows the Planner to forecast GPU load more accurately and handle traffic spikes proactively, supported by routing hints<\/strong> from the Kubernetes Gateway API Inference Extension (GAIE).<\/li>\n<\/ol>\n
    \n

    Check out the\u00a0GitHub Release here<\/a>.\u00a0<\/strong>Also,\u00a0feel free to follow us on\u00a0Twitter<\/mark><\/a><\/strong>\u00a0and don\u2019t forget to join our\u00a0100k+ ML SubReddit<\/a><\/strong>\u00a0and Subscribe to\u00a0our Newsletter<\/a><\/strong>. Wait! are you on telegram?\u00a0now you can join us on telegram as well.<\/a><\/strong><\/p>\n

    The post NVIDIA Releases Dynamo v0.9.0: A Massive Infrastructure Overhaul Featuring FlashIndexer, Multi-Modal Support, and Removed NATS and ETCD<\/a> appeared first on MarkTechPost<\/a>.<\/p>","protected":false},"excerpt":{"rendered":"

    NVIDIA has just released Dynam…<\/p>\n","protected":false},"author":1,"featured_media":29,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-439","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-uncategorized"],"_links":{"self":[{"href":"https:\/\/connectword.dpdns.org\/index.php?rest_route=\/wp\/v2\/posts\/439","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/connectword.dpdns.org\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/connectword.dpdns.org\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/connectword.dpdns.org\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/connectword.dpdns.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=439"}],"version-history":[{"count":0,"href":"https:\/\/connectword.dpdns.org\/index.php?rest_route=\/wp\/v2\/posts\/439\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/connectword.dpdns.org\/index.php?rest_route=\/wp\/v2\/media\/29"}],"wp:attachment":[{"href":"https:\/\/connectword.dpdns.org\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=439"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/connectword.dpdns.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=439"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/connectword.dpdns.org\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=439"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}