Upgrading a drone payload from Xavier NX to Orin NX. How do the Jetson Xavier NX 16GB and Jetson Orin NX 16GB compare on AI performance, memory bandwidth and GPU architecture?

Question

Rapidflare · Accepted Answer

The Jetson Orin NX 16GB gives you more AI compute, more memory bandwidth and a newer GPU than the Jetson Xavier NX 16GB. The Orin NX moves from a Volta GPU to an Ampere GPU. It is rated at up to 100 INT8 TOPS by default and 157 TOPS in Super mode, against 21 TOPS for the Xavier NX. Its memory bandwidth is 102 GB/s, against 59.7 GB/s for the Xavier NX. The two TOPS figures aren't measured the same way, so see the note under the table.

**Jetson Xavier NX 16GB vs. Jetson Orin NX 16GB**

| Spec | Jetson Xavier NX 16GB | Jetson Orin NX 16GB (default) | Jetson Orin NX 16GB (Super mode) |
|---|---|---|---|
| AI performance (INT8) | 21 TOPS (dense) | 100 TOPS (sparse) | 157 TOPS (sparse) |
| FP32 | 0.845 TFLOPS | 1.88 TFLOPS | 2.4 TFLOPS |
| Memory bandwidth | 59.7 GB/s | 102 GB/s | 102 GB/s |
| Memory | 16 GB LPDDR4x | 16 GB LPDDR5 | 16 GB LPDDR5 |
| GPU architecture | Volta | Ampere | Ampere |
| CUDA cores | 384 | 1024 | 1024 |
| Tensor cores | 48 | 32 | 32 |
| NVDLA engines | 2 | 2 | 2 |
| Power envelope | 10–20 W | 10–25 W | 10–40 W |

**Notes on the numbers**
- **The TOPS figures aren't directly comparable.** The Xavier NX's 21 TOPS is a dense INT8 figure, while the Orin NX figures are sparse INT8. The gap in measured throughput on your models may therefore be smaller than the headline numbers suggest.
- **Super mode raises compute and power, not memory bandwidth.** The Orin NX's bandwidth is 102 GB/s in both default and Super mode. Super mode increases INT8 and FP32 throughput, but it also widens the power envelope to as much as 40 W.
- **The GPU is a newer generation with more CUDA cores but fewer Tensor cores.** The Orin NX pairs an Ampere GPU with 1024 CUDA cores and 32 Tensor cores. The Xavier NX has a Volta GPU with 384 CUDA cores and 48 Tensor cores.

**What this means for the drone payload**
- **Mechanical fit:** both modules use the same 260-pin SO-DIMM form factor (69.6x45mm). That doesn't confirm your current carrier board supports the Orin NX, so check carrier compatibility with your board vendor.
- **Power and heat:** the Orin NX's upper power limit is 25 W by default and 40 W in Super mode, against 20 W for the Xavier NX. Size your payload's power supply and cooling for the mode you plan to run.
- **Junction temperature limit:** the Orin NX's maximum is 105 °C, against 90 °C for the Xavier NX. This is a limit for the module's chip, not the outdoor temperature range the drone can fly in.

The [Jetson Orin NX module series data sheet](https://developer.nvidia.com/downloads/jetson-orin-nx-module-series-data-sheet) has the full module specifications.

Jetson Xavier NX 16GB vs Orin NX 16GB: AI performance, memory bandwidth and GPU

Sources

Spec	Jetson Xavier NX 16GB	Jetson Orin NX 16GB (default)	Jetson Orin NX 16GB (Super mode)
AI performance (INT8)	21 TOPS (dense)	100 TOPS (sparse)	157 TOPS (sparse)
FP32	0.845 TFLOPS	1.88 TFLOPS	2.4 TFLOPS
Memory bandwidth	59.7 GB/s	102 GB/s	102 GB/s
Memory	16 GB LPDDR4x	16 GB LPDDR5	16 GB LPDDR5
GPU architecture	Volta	Ampere	Ampere
CUDA cores	384	1024	1024
Tensor cores	48	32	32
NVDLA engines	2	2	2
Power envelope	10–20 W	10–25 W	10–40 W