Selection guide
Jetson Thor vs Orin: Which NVIDIA Jetson Platform to Choose
NVIDIA's Jetson family spans four current platforms. Choose Jetson Thor for humanoids and on-robot foundation models, AGX Orin for multi-camera autonomous machines, Orin NX when size and weight are tight, and Orin Nano for entry-level vision and cost-sensitive products.
Platform comparison
| Platform | AI performance | Memory | Power | Best for |
|---|---|---|---|---|
| Jetson Thor (T5000) | 2070 FP4 TFLOPS (sparse) | 128 GB LPDDR5X, 273 GB/s | 40–130 W | Humanoids and physical AI, running large vision-language and foundation models on the robot, many high-bandwidth sensors |
| Jetson AGX Orin (64GB / 32GB) | 275 / 200 TOPS | 64 GB LPDDR5, 204.8 GB/s | 15–60 W | Multi-camera AMRs, industrial vision, autonomous machines that need 8–16 cameras |
| Jetson Orin NX (16GB / 8GB) | 157 / 117 TOPS | Up to 16 GB LPDDR5, 102.4 GB/s | 10–40 W | Drones, compact robots, and smart cameras where size and weight matter |
| Jetson Orin Nano (8GB / 4GB) | 67 / 34 TOPS | Up to 8 GB LPDDR5, 102 GB/s | 7–25 W | Entry-level vision, prototypes, education, and cost-sensitive volume products |
Hardware for each platform
- Jetson Thor (T5000)
- CTK-5000-A Jetson AGX Thor Developer KitNVIDIA Jetson AGX Thor Developer KitJetson Thor Heatsink — FanJetson Thor Heatsink — Blower
- Jetson AGX Orin (64GB / 32GB)
- COK-800-A Jetson AGX Orin Developer KitNVIDIA Jetson AGX Orin Developer KitCAC-800 Series AGX Orin Carrier BoardCNF-E08 8-Channel GMSL Expansion BoardCNF-E16 16-Channel GMSL Expansion BoardCDH-800 Series Domain ControllerJetson AGX Orin Heatsink
Frequently asked questions
Can I compare Thor's TFLOPS directly with Orin's TOPS?
Not directly. Thor's 2070 figure is FP4 TFLOPS with sparsity, while Orin figures are INT8 TOPS. Both describe peak AI throughput, but at different precisions, so compare them on your own model and precision rather than on the headline number.
Do Orin NX and Orin Nano modules share a carrier board?
Yes. Orin NX and Orin Nano modules use the same form factor and are pin-compatible, so an Orin NX / Orin Nano carrier board such as the CNF-500 or CNC-500 series accepts either module. That lets you prototype on one and ship on the other.
Should I start with an official NVIDIA developer kit or a Haipob kit?
Official NVIDIA kits are the reference for software bring-up and evaluation. Haipob kits package the same modules with our own carrier board, storage, and cooling in a more compact, deployment-ready form, and the carrier board can be adapted for your product.
Does a Jetson module need a heatsink?
Yes. Jetson modules need a thermal solution sized for their power mode. For sustained full-load operation, use active cooling matched to the module's mounting holes and heat-source layout, such as our Thor, AGX Orin, and Orin NX heatsinks.
Which platform should I choose for a humanoid robot?
Jetson Thor. Its 128 GB of memory and Blackwell GPU are designed for running large foundation and vision-language-action models on the robot itself. AGX Orin suits mobile robots with heavy camera workloads but smaller models.
Get a matched recommendation
Share your model size, camera count, interfaces, and power budget. We'll suggest the platform, kit, carrier board, and cooling that fit.