Whether TFLOPS on Tensor Cores is Sparse TFLOPS or DENSE TFLOPS?

Autonomous Machines Jetson & Embedded Systems Jetson AGX Orin
,
1

From the NVIDIA Jetson AGX Orin Series Data Sheet (DS-10662-001_v1.1), there are following sentences.

JAO 64GB: Ampere GPU 2 GPC | 8 TPC | Up to 170 INT8 Sparse TOPS
or 85 FP16 TFLOPS (Tensor Cores) | Up to 5.32 FP32 TFLOPS or
10.649 FP16 TFLOPS (CUDA cores)
JAO 32GB: Ampere GPU 2 GPC | 7 TPC | Up to 108 INT8 Sparse TOPS
or 54 FP16 TFLOPS (Tensor Cores)
Up to 3.365 FP32 TFLOPS or 6.73 FP16 TFLOPS (CUDA cores)

So I want to confirm “85 FP16 TFLOPS (Tensor Cores)” and “54 FP16 TFLOPS (Tensor Cores)” are sparse TFLOPS or dense TFLOPS.

3

85 is Sparse FP16 TFLOPs. Half of this would be Dense FP16 TFLOPs.

1 Like
5

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.