Blue Iris + Deepstack + GPU

[Hans007]

Hi All. i read somewhere that it was recommended to have a GPU installed and use that together with Blue Iris and/or Deepstack, I therefor bought a used (cheap) Nvida Quadro P620, to get better performance.

But sadly it looks worse then before the GPU installed. AI no takes 3649ms where it before was around 1800ms

CPU I5-950 3ghz
RAM 16GB
GPU Nvidia Quadro P620
OS Win10 Pro

I followed Deepstack guide on Windows with GPU and have selected Nvidia NVDEC in Blue Iris. am i missing thing ? or is it just the GPU that is "no-go"

 

[MikeLud1]

Hi All. i read somewhere that it was recommended to have a GPU installed and use that together with Blue Iris and/or Deepstack, I therefor bought a used (cheap) Nvida Quadro P620, to get better performance.

But sadly it looks worse then before the GPU installed. AI no takes 3649ms where it before was around 1800ms

CPU I5-950 3ghz
RAM 16GB
GPU Nvidia Quadro P620
OS Win10 Pro

I followed Deepstack guide on Windows with GPU and have selected Nvidia NVDEC in Blue Iris. am i missing thing ? or is it just the GPU that is "no-go"

Do you have CUDA installed, to check open a command prompt and execute "nvidia-smi" and post the results.

 

[Hans007]

Do you have CUDA installed, to check open a command prompt and execute "nvidia-smi" and post the results.

yes it looks like i do

 

[MikeLud1]

yes it looks like i do
View attachment 138445

Everything looks good, try turning off Nvidia NVDEC your GPU might not have enough memory to do both DeepStack and Nvidia NVDEC decoding

 

[Hans007]

Everything looks good, try turning off Nvidia NVDEC your GPU might not have enough memory to do both DeepStack and Nvidia NVDEC decoding

Will do and get back with the outcome.. thanks so fare.

 

[Hans007]

@MikeLud1,
So just did some more testing, sadly no big of a change. im still getting around 1800-2500ms, any ideas or troubleshooting guide that perhaps can cast some light on why im getting so high of number when using a GPU..

 

[MikeLud1]

@MikeLud1,
So just did some more testing, sadly no big of a change. im still getting around 1800-2500ms, any ideas or troubleshooting guide that perhaps can cast some light on why im getting so high of number when using a GPU..

Which version of deepstack are you using, it should be DeepStack 2022.01.1

 

[Hans007]

Which version of deepstack are you using, it should be DeepStack 2022.01.1

Yes im running the lasted version 2022.01.1

 

[MikeLud1]

Yes im running the lasted version 2022.01.1

Maybe try uninstalling DeepStack and reinstalling, this has worked for some.

 

[Hans007]

Maybe try uninstalling DeepStack and reinstalling, this has worked for some.

okay will give that a try, do i need to remove any trace in regedit before reinstalling it?

 

[MikeLud1]

okay will give that a try, do i need to remove any trace in regedit before reinstalling it?

No

 

[Hans007]

No

did a little more analysis, and what I can see is that in the daytime hours it takes the AI approx 2500ms to identify an object, but it cancels the alert in less then 200ms and at the evening/night hours the AI takes less then 800ms.
and im getting almost the same values on all 6 cameras..

 

[105437]

@MikeLud1 I'm running DS with an NVIDIA P400. I just tried running that command and it wasn't recognized.

 

[MikeLud1]

@MikeLud1 I'm running DS with an NVIDIA P400. I just tried running that command and it wasn't recognized.

Did you install CUDA, nvidia-smi only works if you have CUDA installed

 

[105437]

Did you install CUDA, nvidia-smi only works if you have CUDA installed

I must not have. What version do you recommend for DS and then converting to CodeProject.AI soon?

 

[105437]

I must not have. What version do you recommend for DS and then converting to CodeProject.AI soon?

@MikeLud1 Should I follow the CodeProject.AI instructions and install CUDA based on that even though I'm currently running DS?

 

[MikeLud1]

@MikeLud1 Should I follow the CodeProject.AI instructions and install CUDA based on that even though I'm currently running DS?

When I installed CodeProject.AI I installed the current CUDA version and ran the cuDDN script

 

[105437]

When I installed CodeProject.AI I installed the current CUDA version and ran the cuDDN script

And that won't break DS for me, correct?

 

[MikeLud1]

And that won't break DS for me, correct?

If you are on DeepStack 2022.01.1 it should not break it, older versions of DeepStack it will break it.

 

[105437]

If you are on DeepStack 2022.01.1 it should not break it, older versions of DeepStack it will break it.

Yeah, I'm not on the new version. I'll wait until there's better integration between BI and CPAI and then make the move.