Fiding an elegant solution for an On^4 problem

[gviktora]

You can run nvidia-smi in a loop or if you are using 4.0 you can run it in persistent mode ( nvidia-smi -pm 1)

Thanks for reply Mfatica. The command that goes inside the loop is “nvidia-smi –a –q –l 1”? Thanks.

[mfatica]

nvidia-smi --loop-continuously --interval=60 --filename=/var/log/nvidia-smi.log &

[gviktora]

nvidia-smi --loop-continuously --interval=60 --filename=/var/log/nvidia-smi.log &

Mfatica, this command did not work. Also I tryed another command “nvidia-smi –a –q –l 1”, that did not work too. So I do not know what to do with this context time, it is killing me.

My kernel runs in 170 ms and context time in 12 seconds External Image. My toolkit is 3.2.

[gviktora]

You can run nvidia-smi in a loop or if you are using 4.0 you can run it in persistent mode ( nvidia-smi -pm 1)

Mfatica it worked just fine, I commented the instrution below in my program and there you go.

cudaThreadExit();

Now, the context creation time happens just once. Thank you so much Mfatica.