Jetson AGX Orin 32GB for Edge AI — Specs, Sizing & Deployment Fit
Last updated: August 2026
Datacenter-class AI in an edge form factor: 200 TOPS, 32 GB of unified memory, ~205 GB/s bandwidth, and dual NVDEC engines that carry 8–16 detection streams on one node.
Planning Takeaway
Datacenter-class AI in an edge form factor: 200 TOPS, 32 GB of unified memory, and dual NVDEC engines that carry 8–16 detection streams on one node. At ~$1,799 it is the step between the NX 16GB and the AGX Orin 64GB — and for large 4K deployments EdgeAIStack’s multi-node planner often ranks clusters of these ahead of fewer, bigger nodes.
Specifications
| Spec | Value |
|---|---|
| Compute | 200 INT8 TOPS (Ampere GPU + 2× NVDLA) |
| Memory | 32 GB LPDDR5, ~205 GB/s |
| Power modes | 15W · 30W · 40W · MAXN (15–40 W envelope) |
| Video decode (modeled planning ceiling) | ~13× 1080p / ~5× 4K H.264 concurrent (2× NVDEC) |
| Est. planning capacity | ~16 concurrent 1080p detection streams |
| Cooling | Active required |
| Form factor | Jetson AGX module (100×87 mm) |
| Runtimes / precisions | TensorRT, PyTorch, ONNX, llama.cpp · FP32/FP16/INT8 |
| Software | JetPack 7.2 (Ubuntu 24.04, CUDA 13, June 2026) — current; JetPack 6.x (Ubuntu 22.04) — legacy track |
| Indicative price | ~$1,799 (module, as of mid-2026; AGX Orin 64GB module ~$2,999) |
Sizing Fit
Fits 8–16 camera video-analytics nodes, multi-model pipelines that outgrow 16 GB, and 4K-heavy deployments where its dual NVDEC decode and ~205 GB/s bandwidth are the binding advantage.
- 16-camera smart-city traffic architecture — Reference architecture
- Thor vs AGX Orin decision framework — Comparison
- Multi-model inference on Jetson — Guide
- Jetson lineup power comparison — Benchmarks
- Which VLM fits which Jetson — Guide
- DeepStream vs Frigate: analytics stack choice — Comparison
Decision Framework
Choose the AGX Orin 32GB if:
- You need 8–16 concurrent 1080p streams — or several 4K streams — with detection on a single node.
- Your multi-model pipeline needs more than 16 GB of unified memory but not the 64GB’s VLM headroom.
- You are scaling a 4K deployment across nodes — the multi-node planner often ranks AGX Orin 32GB clusters ahead of fewer, larger nodes on cost and headroom.
Choose a sibling instead if:
- Jetson Orin NX 16GB — your node serves ≤12 streams and models fit in 16 GB (~$999).
- Jetson AGX Orin 64GB — you need ~20 streams, 64 GB memory, or VLM-class models on the same box (~$2,999).
- Jetson T4000 — your roadmap includes FP4-quantized generative models on the current Thor software line (~$2,999).
Frequently Asked Questions
How much does the Jetson AGX Orin 32GB cost?
The AGX Orin 32GB module lists at ~$1,799 as of the July 2026 NVIDIA repricing (up from $899 — a 100% increase); the 64 GB variant lists at ~$2,999. Carrier board and cooling are additional.
How many cameras can the AGX Orin 32GB handle?
EdgeAIStack’s planning model puts it around 16 concurrent 1080p detection streams at 15 FPS, with hardware decode ceilings near ~13× 1080p or ~5× 4K H.264 from its dual NVDEC engines. Real capacity depends on model, resolution, and pipeline — validate with the Hardware Selector.
AGX Orin 32GB vs 64GB — when do I step up?
Step up for VLM-class or generative workloads that need 64 GB of unified memory, or when you need more than ~16 streams (the 64GB is modeled at ~20 with 275 vs 200 TOPS). Both share the same ~205 GB/s bandwidth and dual NVDEC; for 8–16 camera vision work the 32GB at ~$1,799 is usually the better-balanced spend than ~$2,999.
What JetPack version does the AGX Orin 32GB run?
JetPack 7.2 (June 2026) is current for this module — Ubuntu 24.04, Linux kernel 6.8, CUDA 13. JetPack 6.x remains a widely deployed legacy track on Ubuntu 22.04. Moving an existing JetPack 6 install to 7 requires the OTA package or a full reflash — a plain apt upgrade is not an upgrade path.