The module with the most Tensor Cores that works in the PLink-AI 17F1 is the Jetson Xavier NX 16GB. It has 48 Tensor Cores and is rated at 21 TOPS of AI performance (INT8, dense, since it's a Xavier part). Its FP32 performance is 0.845 TFLOPS, and it has no Super mode figure. developer... +1developer.nvidia.com ↗“compute.tensor_cores=48”developer.nvidia.com ↗“performance.max_ai_tops=21 TOPS, performance.max_fp32_tflops=0.845 TFLOPS”
The 17F1 supports four modules: Jetson Xavier NX 16GB, Jetson Orin Nano 8GB, Jetson Orin NX 8GB and Jetson Orin NX 16GB. plink-ai.complink-ai.com ↗“compatibility.jetson_module=Jetson Xavier NX 16GB, Jetson Orin Nano 8GB, Jetson Orin NX 8GB, Jetson Orin NX 16GB”
Supported modules compared nvidia.com +2nvidia.com ↗“compute.tensor_cores=32”nvidia.com ↗“performance.max_ai_tops=70 TOPS”nvidia.com ↗“performance.max_ai_tops=100 TOPS”
| Module | Tensor Cores | AI performance (INT8) | Super mode AI performance (INT8) |
|---|---|---|---|
| Jetson Xavier NX 16GB | 48 | 21 TOPS (dense) | Not listed |
| Jetson Orin Nano 8GB | 32 | 40 TOPS (sparse) | 67 TOPS (sparse) |
| Jetson Orin NX 8GB | 32 | 70 TOPS (sparse) | 117 TOPS (sparse) |
| Jetson Orin NX 16GB | 32 | 100 TOPS (sparse) | 157 TOPS (sparse) |
The Xavier figure is INT8 dense and the Orin figures are INT8 sparse, so they don't compare directly. Even so, the module with the most Tensor Cores has the lowest AI performance of the four. If your goal is maximum AI performance in the 17F1, the Jetson Orin NX 16GB has the highest rating: 100 TOPS by default and 157 TOPS in Super mode, which is the 17F1's "up to 157 TOPS" figure. plink-ai.complink-ai.com ↗“up to 157 TOPS”