What NVIDIA Xavier was
NVIDIA Xavier is the Jetson module generation built around the Xavier system on chip, with an NVIDIA Volta GPU, Tensor Cores, two deep learning accelerators and an 8-core Carmel Arm CPU on the AGX version. It came in two module families: the Jetson AGX Xavier, with up to 32 TOPS, and the smaller Jetson Xavier NX, with up to 21 TOPS. For several years it was the standard choice for vision AI on robots, drones and industrial edge systems.
The Jetson Orin family replaced it. Orin moves to an NVIDIA Ampere GPU, Arm Cortex-A78AE CPU cores and LPDDR5 memory, and NVIDIA lines the modules up so that each Xavier design has a natural Orin successor in a matching form factor. The natural migrations are Xavier NX to Orin NX and AGX Xavier to AGX Orin, while the older TX2 modules step up to the Orin Nano or Orin NX.
Which Orin module replaces each Xavier
The table pairs each Xavier and TX2 module with its usual Orin successor. AI performance is NVIDIA’s peak INT8 figure; Orin figures are quoted with sparsity, so the real-world gain on a given model is smaller than the headline ratio, though still large.
| Xavier-era module | AI performance | Memory | Orin successor | Successor AI performance |
|---|---|---|---|---|
| Jetson Xavier NX | 21 TOPS, 10 to 20 W | 8 or 16 GB LPDDR4x | Jetson Orin NX | Up to 157 TOPS, 10 to 40 W, 8 or 16 GB LPDDR5 |
| Jetson AGX Xavier | 32 TOPS, 10 to 30 W | 32 or 64 GB LPDDR4x | Jetson AGX Orin | Up to 275 TOPS, 15 to 60 W, 32 or 64 GB LPDDR5 |
| Jetson AGX Xavier Industrial | 30 TOPS, 20 to 40 W | 32 GB LPDDR4x with ECC | Jetson AGX Orin Industrial | Up to 248 TOPS, 15 to 75 W, 64 GB with ECC |
| Jetson TX2 NX | 1.33 TFLOPS FP16, 7.5 to 15 W | 4 GB LPDDR4 | Jetson Orin Nano | Up to 67 TOPS, 7 to 25 W, 4 or 8 GB LPDDR5 |
Our AGX Orin vs AGX Xavier comparison looks at the performance step in detail.
Why Xavier is being retired
The Xavier and TX2 modules use LPDDR4 memory, and the 2026 memory shortage brought their end-of-life forward. New orders became non-cancellable and non-returnable, the last-time-buy closed on 1 July 2026, and NVIDIA ships the final modules against placed orders until 15 July 2027. Migrating to Orin is therefore both a performance upgrade and a supply decision. Our Xavier and TX2 end-of-life guide sets out the dates and how to plan around them.
What changes when you migrate
Orin keeps the Xavier form factors, so the move is usually smaller than a full redesign. Several things still need checking before a Xavier system becomes an Orin system.
-
//
Carrier board. Orin NX shares the 260-pin SO-DIMM form factor of the Xavier NX, and AGX Orin shares the 699-pin connector of the AGX Xavier. Power rails, storage and some high-speed lanes differ, so the carrier still needs a design review and often a revision.
-
//
Storage. The Orin NX and Orin Nano modules have no onboard eMMC, so the carrier needs an NVMe SSD or other boot storage.
-
//
Software. Xavier tops out at JetPack 5. Orin runs JetPack 6 and later, so the application is rebuilt on the newer CUDA, TensorRT and Ubuntu release and models are re-optimised.
-
//
Thermal design. Orin modules can draw more power at their top modes, so the heatsink and enclosure are checked against the chosen power mode.
-
//
Certification. A new module in a certified product, such as rail or medical equipment, usually means some retesting.
Planning a Xavier to Orin migration
Most programs run two tracks at once. The first covers the systems already in the field: how many Xavier modules are needed for remaining production, repairs and spares until the Orin version is approved. The second is the Orin design itself, which starts with choosing the module from the real workload rather than the headline TOPS figure.
A practical order of work is to benchmark the existing model on the target Orin module, confirm the carrier board and storage, rebuild the software image on the current JetPack release, then run thermal and environmental checks in the final enclosure. Keeping the same camera, I/O and mounting arrangement where possible limits the change for the end customer and for any field engineers who maintain the system.
Choosing between Orin and Thor
For most Xavier designs, the matching Orin module is the right step: the same class of system, a lower-risk migration and a large performance gain. Our Jetson Orin NX computers take over from Xavier NX systems, and our Jetson AGX Orin computers from AGX Xavier systems. Where the application is moving to large language or vision-language models, the newer Jetson Thor modules may be the better target, at a higher power envelope. Existing Jetson Xavier systems remain available while stock lasts for repair and spares.
Quick answers on NVIDIA Xavier
NVIDIA Xavier: common questions
Is the Jetson Xavier end of life?
Yes. The Xavier NX, AGX Xavier and TX2 modules are end-of-life. The last-time-buy closed on 1 July 2026, with final shipments against placed orders running to 15 July 2027.
Is the Orin NX pin-compatible with the Xavier NX?
Largely, but check the carrier. Both use the 260-pin SO-DIMM form factor and NVIDIA designed the Orin NX as the successor, but power, boot storage and some interfaces differ, so a carrier design review is needed.
Which JetPack supports Xavier?
JetPack 5. It is the last major release for Xavier. JetPack 6 and later support Orin and newer modules only.
What replaced the Jetson TX2?
The Jetson Orin Nano, or the Orin NX for heavier workloads. Both deliver far more AI performance than the TX2 or TX2 NX at similar power.
Is Orin much faster than Xavier?
Yes. NVIDIA quotes up to 8 times the AI performance for AGX Orin over AGX Xavier. Gains on a real model are smaller than the sparse peak figure suggests, but still several times faster.
We manage Xavier to Orin migrations as standard: sizing any remaining stock of the old module, selecting the Orin module and carrier, fitting storage and the JetPack image, and validating the new system before it ships.