Can I calibrate my network on the old graphics card with low compute capability?

Hi NVIDIA,

I want to do calibration with TensorRT, so I tried the example tf-trt. I run the example on two graphics card: Tesla K40 and GTX970m. The log show that both card doesn’t support INT8 and FP16, but it still work and output the pb file. The pb file is larger than I expected.

Follow is the size of pb file and the Tensorflow/TensorRT Environment:

The sample doesn’t work correctly and I guess that it may related with my graphics card.

From follow page I find that FP16 and INT8 only support cards with compute capability above 6.1.
https://devtalk.nvidia.com/default/topic/1023708/gpu-accelerated-libraries/fp16-support-on-gtx-1060-and-1080/
Does it mean I can’t do calibration on my old graphics card? Or they just work slower than card with high cc.

I get sample from this page: https://devblogs.nvidia.com/tensorrt-integration-speeds-tensorflow-inference/

Best Regards
Matthias

Hello,

The following table lists NVIDIA hardware and which precision modes each hardware supports. It also lists availability of Deep Learning Accelerator (DLA) on these hardware. TensorRT supports all NVIDIA hardware with capability SM 3.0 or higher.

https://docs.nvidia.com/deeplearning/sdk/tensorrt-support-matrix/index.html#hardware-precision-matrix