Skip to main content

HyperPod Inference Operator (Managed KV Cache and Intelligent Routing)

Published 2026-06-23Updated 2026-06-3012 min read

Overview

Amazon SageMaker HyperPod Inference Operator is a managed component that serves LLM inference on EKS. It provides Managed Tiered KV Cache and Intelligent Routing as an EKS add-on, reducing prefill recomputation and improving throughput for vLLM-based workloads.

This document describes which layer of the Tiered Gateway Architecture HyperPod's inference routing maps to, what differs from a full-stack gateway, and the configuration of KV cache, routing strategies, and instance requirements. The intended audience is platform engineers evaluating and configuring HyperPod inference endpoints.

Background: Two Routing Layers of HyperPod Inference

A HyperPod inference endpoint consists of two routing layers. This distinction is the starting point for the question "is the HyperPod gateway a full-stack gateway?"

LayerComponentResponsibilitiesNot Responsible For
L1 (Edge)Application Load BalancerTLS termination, health checks, DNS (optional Route 53 integration)AuthN/AuthZ, rate limiting, model/provider selection
L2 (Inference Routing)Intelligent Router (Inference Operator)KV cache state-based Pod selection (prefix-aware, etc.)AuthN, rate limiting, context-aware (semantic) routing

Per-request rate limiting is handled not by the router but by a per-Pod nginx sidecar (the InferenceEndpointConfig RequestLimits: maxConcurrentRequests, maxQueueSize, overflowStatusCode). This is Pod-level (L3) control, not gateway-level policy.

HyperPod Intelligent Router vs Tiered Gateway

The 2-Tier gateway in the Routing Strategy document separates the L1 edge gateway (AuthN, rate limit, TLS) and L2 inference routing (KV-aware Pod selection) into distinct components. The HyperPod Intelligent Router handles only L2 of these.

FeatureHyperPod Intelligent RouterTiered Gateway (kgateway + EPP)
KV cache-aware Pod routingProvided (prefixaware · kvaware, etc.)Provided (EPP prefix-cache-scorer)
Management modelAWS managed (EKS add-on)Self-managed (K8s standard)
AuthN/AuthZNot provided (configure separately upstream)Configured at the gateway level
Rate limitingPer-Pod nginx sidecar (L3)Token/request-based at the gateway level
Context-aware (semantic) routingNot providedIntegrated with LLM Classifier · vLLM Semantic Router
MCP/A2ANot providedIntegrated with agentgateway
Key distinction

With HyperPod, the ALB (L1) and Intelligent Router (L2) are deployed together and appear architecturally co-located, but full-stack gateway features such as rate limiting, authentication, and semantic routing are not included. If these features are needed, a separate L1 gateway (kgateway, Kong, etc.) must be placed in front of HyperPod. In a self-managed EKS environment, the gateway and router are separate components from the start.

KV Cache Configuration: L1/L2 Cache and Routing Strategies

HyperPod's managed KV cache is configured as a 2-tier cache in the inference endpoint settings.

  • L1 Cache: CPU memory on each inference node. Local low-latency reuse on the node.
  • L2 Cache: A layer shared across nodes. Choose redis (customer-managed) or tieredstorage (HyperPod-managed distributed memory) as the backend.

Configuration is performed in the KV cache spec of InferenceEndpointConfig by enabling enableL1Cache and enableL2Cache and specifying l2CacheBackend.

# Conceptual example — verify exact field names and schema against the Operator version in use
kvCacheSpec:
enableL1Cache: true
enableL2Cache: true
l2CacheSpec:
l2CacheBackend: tieredstorage # or redis

Four Routing Strategies

Intelligent Routing sends incoming requests to the instance most likely to hold the relevant KV cache. The following strategies are provided.

StrategyBehavior
prefixaware (default)Routes requests with the same prompt prefix to the same instance
kvawareRoutes to the instance with the highest KV cache hit rate
sessionRoutes requests from the same user session to the same instance
roundrobinDistributes evenly without considering KV cache state

The kvaware strategy has constraints. It works only with vLLM-based images, requires the /completions endpoint (/v1/chat/completions not supported), and needs LMCache/vLLM versions compatible with Inference Operator v3.1.3 or later.

P-type instance requirement needs verification (unconfirmed)

The requirement that managed tiered KV cache (L2 tieredstorage) is created only on P-series instances has not been confirmed in the official documentation. The part where official documentation explicitly requires an instance family is the Disaggregated Prefill/Decode (DPD) below, which requires EFA · GPUDirect RDMA-capable instances (ml.p5 · ml.p5e · ml.p5en · ml.p6-b200 · ml.p6-b300).

In the field, there have been reports that "the tiered KV cache configuration object is not created on non-P-type instances," but this is not reflected in public documentation. The exact question — including "since which version is it P-only" — is an open item that must be directly verified against the Operator version in use and the actual CRD spec. The object name (TieredKvcacheConfig, etc.) may also differ across versions, so re-checking the release notes at the time of application is recommended.

Disaggregated Prefill/Decode (DPD)

HyperPod introduced managed Disaggregated Prefill/Decode in v3.2 (2026-06). This is a pattern that separates prefill (compute-bound) and decode (memory-bound) so each stage can scale independently, with a strong effect on long contexts and large models (see Disaggregated Serving for the concept).

DPD requires EFA · GPUDirect RDMA-capable instances for KV cache transfer: ml.p5.48xlarge · ml.p5e.48xlarge · ml.p5en.48xlarge · ml.p6-b200.48xlarge · ml.p6-b300.48xlarge. For workloads with short, repetitive prompts such as search-result summarization, prefix cache reuse (see the throughput levers below) provides a more direct gain than DPD.

Throughput (TTFT · TPS) Levers

In workloads such as search summarization where the same keywords repeat and KV cache hit rate dominates throughput, the following levers have a direct effect. For numbers and detailed configuration, see KV Cache Optimization.

LeverEffectIn HyperPod
Prefix cache reuseTTFT 50–80%↓ for the same system prompt, throughput 400%+↑prefixaware / kvaware routing + L2 tiered cache
Automatic Prefix Caching (vLLM)Skip prefill for repeated prefixesvLLM container --enable-prefix-caching
Chunked PrefillBalance TTFT and throughputvLLM engine option
DPDImproves tail latency for long contextsv3.2 managed (requires EFA P5/P6)

The routing decision (prefix hash lookup) is not inference; only the final workload is inference. For this distinction, see the routing note in the KV Cache Optimization document.

Fit and Limitations

AspectDetails
Good fitMinimal ops staffing, automatic node recovery and governance, vLLM-based serving, leveraging AWS-managed KV cache
Backend constraintsvLLM only. kvaware only /completions. If TensorRT-LLM performance ceiling is required, consider self-managed options such as Dynamo
Gateway featuresAuthN, rate limiting, semantic routing, and MCP not included → configure a separate L1 gateway upstream
Cost+15–20% instance premium over EC2 (e.g., ml.p5 $66 vs p5 $55, us-east-1, 2026-06). Auto-recovery and governance can raise utilization and offset this
AvailabilityVerify region and instance-family availability. P-series often has supply constraints

Managed KV cache and intelligent routing went GA in 2025-11, and DPD was introduced in v3.2 (2026-06). Features and versions should be reconfirmed against the HyperPod inference release notes at the time of application.

References

Official Documentation