The Jetson Orin Nano 8GB has more AI TOPS: 40 TOPS in default mode and 67 TOPS in Super mode, against 21 TOPS for the Jetson Xavier NX 8GB. The Xavier NX 8GB has more Tensor cores, though: 48 versus 32.
Jetson Orin Nano 8GB vs Jetson Xavier NX 8GB nvidia.com +1nvidia.com ↗“performance.max_ai_tops=40 TOPS, performance.max_ai_tops_super=67 TOPS”developer.nvidia.com ↗“performance.max_ai_tops=21 TOPS”
| Spec | Jetson Orin Nano 8GB | Jetson Xavier NX 8GB |
|---|---|---|
| AI TOPS (default mode) | 40 TOPS (INT8, sparse) | 21 TOPS (INT8, dense) |
| AI TOPS (Super mode) | 67 TOPS (INT8, sparse) | No Super figure |
| Power envelope (default mode) | 7–15 W | 10–20 W |
| Power envelope (Super mode) | 7–25 W | No Super figure |
| Tensor cores | 32 | 48 |
Notes:
- The TOPS numbers aren't measured the same way. Both are INT8, but the Orin figure counts sparse operations and the Xavier figure counts dense operations. This means the two numbers can't be compared one-to-one.
- Super mode only applies to the Orin Nano. The Xavier NX has no Super-mode figures.
- Two different Xavier NX power ranges are listed. The detailed spec says 10–20 W, but the short product description says 10–15 W. I can't tell which applies, so check it against NVIDIA's datasheet before planning your power budget. developer... +1developer.nvidia.com ↗“power.power_envelope=10-20 W”developer.nvidia.com ↗“8GB LPDDR4x, 10-15W”
- The Tensor cores are from different GPU generations. The Orin Nano's GPU is Ampere and the Xavier NX's is Volta. Because of that, the core counts alone don't show which module runs AI workloads faster. nvidia.com +1nvidia.com ↗“compute.gpu_architecture=Ampere”developer.nvidia.com ↗“compute.gpu_architecture=Volta”