How to use GPU and CPU at the same time (TX2)


is there a possibility to use CPU and GPU at once at the Nvidia Jeton TX2 maybe via Blas routines or something else, especially i want to use for caffe or tensorrt cpu and gpu to get the best speed up.

Best regards.

Not sure what do you mean by “to use CPU and GPU at once”…when you use cuBlas, it’s accelerated by GPU. Same for cuDNN or TensorRT 3.0. There is certainly CPU implementation for their corresponding functions, such as BLAS. Should there is a need to use CPU version, you can. Is this for benchmark comparison?

I run imagenet-console with TensorRT in jeston tx2, it spends about 30s to classify 300 pictures。

How can I confirm TensorRT using GPU or not?

Is there a CPU counterpart version you could compare with under the same condition/configuration? More info on your test code that we can use to repro your result? Yes, TRT is an inference optimizer and supposed to speed up the operation.

Ok thx for your answers.

Best regards.

TensorRT is really fast, but most of the time comes from initialize.

Best regards.