Dell r80 with Blue Iris and 3 cameras drive CPU 100% Max need to add Hardware GPU

[devoir]

Setting up a Dell r820 server running XCP-ng for multiple VM's I have a VM setup with Windows 10 and the latest Blue Iris update.

When I try to run more that 3 cameras the 15 CPU's max out at 100%. Per the help file you can add a GPU to aid in hardware processing.

These were to GPU choices. https://developer.nvidia.com/video-e...support-matrix


I installed a Nvidia Qudra 600, following the help guide to set camara record direct to disc and on each camera chose Intel or Nviida for the Hardware accelerator

Despite this I can't find a setting to reduce the cpus load below 75%, I need to add about 10 more cameras.

If you have any experience with this I would appreciate your guidance and assistance


Devoir

[Fellhahn]

Per camera, in the camera settings, on the video tab. There is a checkbox "limit decoding unless required". Is that ticked for each camera?

What you're experiencing doesn't sound normal. I have a Intel Xeon E-2146, which is a hexa-core chip with HT circa 2019. Ubuntu 20.04 host, KVM/QEMU guest running win10.
6 vCPUs/threads (whatever you want to call them) are made available to the guest. I have no GPU passed through (though am planning on buying a quadro at some point).

I have 6 cameras in my Blue Iris setup streaming 1440p H.265. CPU flops around between 40% and 60%.

[IanIreland]

If you go to the IPCamTalk forum there is guidance on reducing the CPU load.

[Eman]

You may find some info here :https://ipcamtalk.com/wiki/optimizin...s-s-cpu-usage/



Eman.

[Rvtravlr]

Limit decoding is not recommended if you are using BI for motion detection. To reduce CPU load it is recommended to use sub streams.

[Fellhahn]

Limit decoding is not recommended if you are using BI for motion detection. To reduce CPU load it is recommended to use sub streams.

Interesting, so set up an additional group of cameras for the sub-streams, as hidden cameras. Decode the sub stream full time, perform motion detection on the sub stream, and have alerts trigger the full res camera as well?

[Fellhahn]

I've also been interested in following the guide provided here for "AI" person detection done in house.

https://www.youtube.com/watch?v=fwoonl5JKgo

[Rvtravlr]

I've also been interested in following the guide provided here for "AI" person detection done in house.

https://www.youtube.com/watch?v=fwoonl5JKgo

I have some cameras using in house AI detection and some using Sentry AI, a paid, cloud service. I like both but prefer the in house.

Go over to ipcamtalk.com and look under Blue Iris, Customizing. There is also a wiki that helps you to set up Blue Iris.

[Tomgru]

I have some cameras using in house AI detection and some using Sentry AI, a paid, cloud service. I like both but prefer the in house.

Go over to ipcamtalk.com and look under Blue Iris, Customizing. There is also a wiki that helps you to set up Blue Iris.

Ipcamtalk +1.

Also. If I remember correctly, VMs aren't great for BI. I ended up removing the VM and put both BI and HS on the same server and my CPU load went way down. No problems since.


Sent from my SM-G975U using Tapatalk

[Rvtravlr]

Interesting, so set up an additional group of cameras for the sub-streams, as hidden cameras. Decode the sub stream full time, perform motion detection on the sub stream, and have alerts trigger the full res camera as well?

No need for cloned cameras. Blue Iris can accept both streams from a camera.

[Fellhahn]

While this conversation has been going on I've also tried something else new, isolating CPUs from host usage and pinning them to vCPUs for a KVM/QEMU guest.

I've dropped back from 6 threads/cores given to the guest down to 4. Despite the reduction CPU usage actually dropped by around 15% at "idle".

Followed this guide here:

Performance Improvements in VMs by adjusting CPU pinning and assignment - VM Engine (KVM) - Unraid

I don't use Unraid but the changes are still applicable.

Huge Pages are next on my list of things to wrap my head around.


Hoping to see something like a Quadro RTX 600 or 1000 from Nvidia soon. Currently the lowest end Turing based Quadro is the 4000, still way too pricey for me.

[Dweber85rc]

What is your disk system like? If your disks are not fast enough then that will mask itself with increased cpu and memory to make up for the lack of disk speed in a virtual environment. You also don't mention what virtualization platform you are using and while yes some of the principles of machine configuration are the same across the different platforms , it isn't true for all settings. I have my BI system running on a VM in hyperv and while I only have 3 cameras, with 8 vcpu assigned it never gets above 15% inside the vm. But I also manage several esxi stretched metro clusters at work and what works in esxi does not always work in hyperv