gpu decoding vs cpu decoding for libtorch

gpu decoding for libtorch is slow than cpu decoding on a same audio ， whether the speed of my test on cpu and gpu is normal？

gpu decode：
real    1m5.747s
user    0m47.806s
sys     0m16.468s
cpu decode：
real    0m39.771s
user    0m37.880s
sys     0m1.975s

what is the chunk size in your testing?

I use default chunk size， both for cpu and gpu decoding.

I modify the chunk size as “-1、1、10” in params.h，but the decoding speed had no change. @robin1001

Exactly the same problem. did you solve it?

Exactly the same problem. did you solve it?

no，move to another framework.

Thanks for the reply!

please try to repeat 10 times

This might be a warmup issue

This issue has been automatically closed due to inactivity.