Jetson Orin Nano 8GB uses the newer Ampere GPU architecture, while Jetson Nano uses Maxwell. Orin Nano 8GB also has 8 times as many CUDA cores (1024 vs. 128) and twice the memory (8 GB LPDDR5 vs. 4 GB LPDDR4), with much higher memory bandwidth.
Jetson Nano vs. Jetson Orin Nano 8GB developer... +1developer.nvidia.com ↗“Entry Maxwell SoM; 472 GFLOPS FP16, 128 CUDA cores, 4-core Cortex-A57, 4GB LPDDR4”nvidia.com ↗“Entry Ampere SoM; up to 67 INT8 TOPS (Super mode), 1024 CUDA + 32 Tensor cores”
| Attribute | Jetson Nano | Jetson Orin Nano 8GB |
|---|---|---|
| GPU architecture | Maxwell | Ampere |
| CUDA cores | 128 | 1024 |
| Tensor cores | 0 | 32 |
| Memory capacity | 4 GB | 8 GB |
| Memory type | LPDDR4 | LPDDR5 |
| Memory bandwidth | 25.6 GB/s | 68 GB/s (default), 102 GB/s (Super mode) |
How to read the memory bandwidth figures:
- Jetson Orin Nano 8GB has two bandwidth figures: 68 GB/s in default mode and 102 GB/s in Super mode. nvidia.comnvidia.com ↗“8GB LPDDR5 (102 GB/s), 7-25W”
- Jetson Nano has a single figure of 25.6 GB/s. developer...developer.nvidia.com ↗“4GB LPDDR4 (25.6 GB/s), 5-10W”
You didn't ask about tensor cores, but I included them because they're part of the GPU difference.