Compatibility | NVIDIA Dynamo Documentation

Compatibility by Version

Dynamo v1.3.1Patch

Dynamo version

Released Aug 5, 2026 · Release notes · UCX 1.20.x

SGLang0.5.14

NIXL1.3.2

CUDA 13.0Driver 580.xx+

TensorRT-LLM1.3.0rc19

NIXL1.0.1

CUDA 13.1Driver 580.xx+

vLLM0.23.0

NIXL1.1.0

CUDA 13.0Driver 580.xx+

GPU

BlackwellHopperAda LovelaceAmpere

Ubuntu 24.04Ubuntu 22.04 · wheels only

Arch

x86_64ARM64 (Ubuntu 24.04 only)

Patch release. Fixes disaggregated SGLang serving over AWS EFA on GB200: the SGLang EFA runtime moves to NIXL 1.3.2 and all three EFA images to EFA Installer 1.49.0. Backend pins are otherwise unchanged from v1.3.0.

These are the backend versions tested and supported for this release. TensorRT-LLM does not support Python 3.11.

The card above covers one release at a time. To compare CUDA toolkit and minimum driver requirements across releases per backend, jump to the Release Support Matrix at the bottom of this page. For extended driver compatibility beyond the listed minimums, including forward compatibility and cuda-compat packages, see the CUDA Compatibility documentation.

See Release Artifacts for the full artifact inventory — container images, wheels, Helm charts, and crates — Local Installation for host OS and architecture requirements, and Model Early Access Builds for per-model early access container builds. For backend-specific runtime workarounds — including the local docker run --network host hang first reported on Amazon Linux 2023 — see TensorRT-LLM Known Issues.

Feature Support

Feature support by backend

SupportedCaveatExperimentalNot supported

SGLang9 / 15

TRT-LLM9 / 15

vLLM14 / 15

Disaggregated Serving

KV-Aware Routing

SLA-Based Planner

KV Block Manager

Multimodal (Image)

Multimodal (Video)

—

Multimodal (Audio)

—

Request Migration

Request Cancellation

LoRA

—

Tool Calling

Speculative Decoding

GPU Memory Service

Shadow Engine Failover

Dynamo Snapshot

Hover a noted cell or focus it with the keyboard to view its compatibility note.

Per-Backend Detail

vLLM

SGLang

TensorRT-LLM

vLLM offers the broadest feature coverage in Dynamo, with full support for disaggregated serving, KV-aware routing, KV block management, LoRA adapters, and multimodal inference including video and audio.

Source: docs/backends/vllm/README.md

Feature	Supported?	Notes
Disaggregated Serving	✓	Prefill/decode separation with NIXL KV transfer
KV-Aware Routing	✓
SLA-Based Planner	✓
KV Block Manager	✓
Multimodal	✓	Image + video; audio experimental (Qwen2-Audio). With KV-aware routing, image-aware routing on documented paths (Source)
Request Migration	✓
Request Cancellation	✓
LoRA	✓	Dynamic load/unload; KV-aware routing supports adapter affinity
Tool Calling	✓
Speculative Decoding	✓	Eagle3 (Source)
GPU Memory Service	✓	Weights and KV; upstream integration remains in progress
Shadow Engine Failover	!	Software-process failover only; no KV-cache reuse or hardware fault tolerance
Dynamo Snapshot	!	Single-GPU supported; multi-GPU is highly experimental and multinode remains in progress

Feature Interactions

Pairwise feature-by-feature compatibility within each backend. Each cell reports whether the row feature works together with the column feature.

vLLM

SGLang

TensorRT-LLM

vLLM Feature Interactions

SupportedExperimentalNot supportedNot applicable

	Disaggregated Serving	KV-Aware Routing	SLA-Based Planner	KV Block Manager	Multimodal	Request Migration	Request Cancellation	LoRA	Tool Calling	Speculative Decoding
Disaggregated Serving
KV-Aware Routing
SLA-Based Planner
KV Block Manager
Multimodal
Request Migration
Request Cancellation
LoRA
Tool Calling
Speculative Decoding

Each cell reports whether the row feature works together with the column feature. Blank cells mirror the populated lower triangle. Hover a noted cell or focus it with the keyboard to view its note.

Release Support Matrix

Every stable release and patch, grouped by minor line — newest first. Expand a line to see each release’s backend pins, CUDA toolkit, and minimum driver, ordered CUDA 12 before CUDA 13. Releases predating per-release CUDA tracking are listed with their requirements marked “Not recorded” rather than dropped. Platform previews and model-specific builds are excluded; the notes below call out the ones whose toolkit support differs, and the Releases (machine-readable) page has the full inventory.

1.3.x2 releasesCUDA 13

Release	Backend	Engine	CUDA toolkit	Min driver	NIXL	Note
v1.3.1patch · Aug 5, 2026 UCX 1.20.x	SGLang	0.5.14	13.0	580.xx+	1.3.2	—
	TensorRT-LLM	1.3.0rc19	13.1	580.xx+	1.0.1	—
	vLLM	0.23.0	13.0	580.xx+	1.1.0	—
v1.3.0stable · Jul 20, 2026 UCX 1.20.x	SGLang	0.5.14	13.0	580.xx+	1.3.0	—
	TensorRT-LLM	1.3.0rc19	13.1	580.xx+	1.0.1	—
	vLLM	0.23.0	13.0	580.xx+	1.1.0	—

1.2.x2 releasesCUDA 12, 13

Release	Backend	Engine	CUDA toolkit	Min driver	NIXL	Note
v1.2.1patch · Jun 13, 2026 UCX not specified	SGLang	0.5.11	12.9	575.xx+	1.0.1	—
	vLLM	0.20.1	12.9	575.xx+	0.10.1	—
	SGLang	0.5.11	13.0	580.xx+	1.0.1	—
	TensorRT-LLM	1.3.0rc14	13.1	580.xx+	0.10.1	—
	vLLM	0.20.1	13.0	580.xx+	0.10.1	—
v1.2.0stable · Jun 2, 2026 UCX 1.20.0	SGLang	0.5.11	12.9	575.xx+	1.0.1	—
	vLLM	0.20.1	12.9	575.xx+	0.10.1	—
	SGLang	0.5.11	13.0	580.xx+	1.0.1	—
	TensorRT-LLM	1.3.0rc14	13.1	580.xx+	0.10.1	—
	vLLM	0.20.1	13.0	580.xx+	0.10.1	—

1.1.x2 releasesCUDA 12, 13

Release	Backend	Engine	CUDA toolkit	Min driver	NIXL	Note
v1.1.1patch · May 5, 2026 UCX not specified	SGLang	0.5.10.post1	12.9	575.xx+	1.0.1	—
	vLLM	0.19.0	12.9	575.xx+	0.10.1	—
	SGLang	0.5.10.post1	13.0	580.xx+	1.0.1	—
	TensorRT-LLM	1.3.0rc11	13.1	580.xx+	0.10.1	—
	vLLM	0.19.0	13.0	580.xx+	0.10.1	—
v1.1.0stable · May 1, 2026 UCX 1.20	SGLang	0.5.10.post1	12.9	575.xx+	1.0.1	—
	vLLM	0.19.0	12.9	575.xx+	0.10.1	—
	SGLang	0.5.10.post1	13.0	580.xx+	1.0.1	—
	TensorRT-LLM	1.3.0rc11	13.1	580.xx+	0.10.1	—
	vLLM	0.19.0	13.0	580.xx+	0.10.1	—

1.0.x3 releasesCUDA 12, 13

Release	Backend	Engine	CUDA toolkit	Min driver	NIXL	Note
v1.0.2patch · Apr 22, 2026 UCX not specified	SGLang	0.5.9	12.9	575.xx+	0.10.1	—
	vLLM	0.16.0	12.9	575.xx+	0.10.1	—
	SGLang	0.5.9	13.0	580.xx+	0.10.1	—
	TensorRT-LLM	1.3.0rc5.post1	13.1	580.xx+	0.10.1	—
	vLLM	0.16.0	13.0	580.xx+	0.10.1	—
v1.0.1patch · Mar 16, 2026 UCX not specified	SGLang	0.5.9	12.9	575.xx+	0.10.1	—
	vLLM	0.16.0	12.9	575.xx+	0.10.1	—
	SGLang	0.5.9	13.0	580.xx+	0.10.1	—
	TensorRT-LLM	1.3.0rc5.post1	13.1	580.xx+	0.10.1	—
	vLLM	0.16.0	13.0	580.xx+	0.10.1	—
v1.0.0stable · Mar 12, 2026 UCX not specified	SGLang	0.5.9	12.9	575.xx+	0.10.1	—
	vLLM	0.16.0	12.9	575.xx+	0.10.1	—
	SGLang	0.5.9	13.0	580.xx+	0.10.1	—
	TensorRT-LLM	1.3.0rc5.post1	13.1	580.xx+	0.10.1	—
	vLLM	0.16.0	13.0	580.xx+	0.10.1	—

0.9.x2 releasesCUDA 12, 13

Release	Backend	Engine	CUDA toolkit	Min driver	NIXL	Note
v0.9.1patch · Mar 4, 2026 UCX not specified	SGLang	0.5.8	12.9	575.xx+	0.9.0	—
	vLLM	0.14.1	12.9	575.xx+	0.9.0	—
	TensorRT-LLM	1.3.0rc3	13.0	580.xx+	0.9.0	—
v0.9.0stable · Feb 11, 2026 UCX not specified	SGLang	0.5.8	12.9	575.xx+	0.9.0	—
	vLLM	0.14.1	12.9	575.xx+	0.9.0	—
	TensorRT-LLM	1.3.0rc1	13.0	580.xx+	0.9.0	—

0.8.x5 releasesCUDA 12, 13

Release	Backend	Engine	CUDA toolkit	Min driver	NIXL	Note
v0.8.1patch · Jan 23, 2026 UCX not specified Post trains v0.8.1.post3, v0.8.1.post2, v0.8.1.post1 share this CUDA support	SGLang	0.5.6.post2	12.9	575.xx+	0.8.0	—
	vLLM	0.12.0	12.9	575.xx+	0.8.0	—
	SGLang	0.5.6.post2	13.0	580.xx+	0.8.0	Experimental
	TensorRT-LLM	1.2.0rc6.post1	13.0	580.xx+	0.8.0	—
	vLLM	0.12.0	13.0	580.xx+	0.8.0	Experimental
v0.8.0stable · Jan 15, 2026 UCX not specified	SGLang	0.5.6.post2	12.9	575.xx+	0.8.0	—
	vLLM	0.12.0	12.9	575.xx+	0.8.0	—
	SGLang	0.5.6.post2	13.0	580.xx+	0.8.0	Experimental
	TensorRT-LLM	1.2.0rc6.post1	13.0	580.xx+	0.8.0	—
	vLLM	0.12.0	13.0	580.xx+	0.8.0	Experimental

0.7.x3 releasesCUDA 12, 13

Release	Backend	Engine	CUDA toolkit	Min driver	NIXL	Note
v0.7.1patch · Dec 15, 2025 UCX not specified	SGLang	0.5.4.post3	12.8	570.xx+	0.8.0	—
	vLLM	0.11.0	12.9	575.xx+	0.8.0	—
	TensorRT-LLM	1.2.0rc3	13.0	580.xx+	0.8.0	—
v0.7.0stable · Nov 26, 2025 UCX not specified Post train v0.7.0.post1 shares this CUDA support	SGLang	0.5.4.post3	12.9	575.xx+	0.8.0	—
	vLLM	0.11.0	12.8	570.xx+	0.8.0	—
	TensorRT-LLM	1.2.0rc2	13.0	580.xx+	0.8.0	—

0.6.x3 releasesCUDA not recorded

Release	Backend	Engine	CUDA toolkit	Min driver	NIXL	Note
v0.6.1patch · Nov 6, 2025 UCX not specified Post train v0.6.1.post1 shares this CUDA support	SGLang	0.5.3.post2	Not recorded	Not recorded	0.6.0	—
	TensorRT-LLM	1.1.0rc5	Not recorded	Not recorded	0.6.0	—
	vLLM	0.11.0	Not recorded	Not recorded	0.6.0	—
v0.6.0stable · Oct 28, 2025 UCX not specified	SGLang	0.5.3.post2	Not recorded	Not recorded	0.6.0	—
	TensorRT-LLM	1.1.0rc5	Not recorded	Not recorded	0.6.0	—
	vLLM	0.11.0	Not recorded	Not recorded	0.6.0	—

Patch versions (e.g. v0.8.1.post1, v0.7.0.post1) have the same CUDA support as their base version.
Early access v1.1.0-dev.* images follow the same CUDA matrix as v1.0.2. The v1.2.0-deepseek-v4-dev.3 vLLM container is CUDA 13.0 multi-arch; the SGLang containers split by arch (CUDA 12.9 on amd64, CUDA 13.0 on arm64).
Experimental CUDA 13 images are not published for all versions.

Driver already installed? Read across from your version — each cell is the newest release that backend can run on it. A driver meeting a higher floor also runs everything below it.

Driver	SGLang	TensorRT-LLM	vLLM
570.xx+	0.7.1	None	0.7.0.post1
575.xx+	1.2.1	None	1.2.1
580.xx+	1.3.1	1.3.1	1.3.1