Blue Iris AI an NVIDIA 3070

wpiman

Getting comfortable
Jul 16, 2018
474
334
massachusetts
I am running Blue Iris 6 with the embedded AI on a Windows Proxmox machine. I have a GeForce 3060 passed through and It was working with codeproject. I can select it in Blue Iris (there are two instance in each). When I get motion, the AI appears to work-- but my CPU usage is skyrocketting. I am seeing the GPU barely move... Like here it is at 7%. It sits idle beween 3% and 6%.

God granted this hardware is getting a little old... but I am wondering if I am setting something up wrong and how I would debug this?



Screenshot 2026-03-21 at 5.14.43 PM.png
Screenshot 2026-03-21 at 5.14.23 PM.png
 
I ran your issue through ChatGPT and here's the response...

With the v6 built-in AI in Blue Iris, seeing two identical entries for your NVIDIA GeForce RTX 3060 is expected behavior and comes down to how Blue Iris exposes GPU acceleration internally.

What’s actually happening​

In v6, Blue Iris uses newer Windows AI/GPU APIs, and your RTX 3060 is exposed through multiple execution providers, typically:
  • DirectML (DML) → Microsoft’s universal GPU AI layer
  • CUDA (sometimes indirectly exposed) → NVIDIA’s native compute path
Even though these are different backends, Blue Iris doesn’t label them separately, so you end up seeing:
  • NVIDIA GeForce RTX 3060
  • NVIDIA GeForce RTX 3060
Same hardware, different AI pipelines.

Why this is more noticeable in v6​

Earlier setups (with CodeProject.AI or DeepStack) made backend differences clearer.

In v6 built-in AI:
  • Blue Iris handles inference internally
  • It queries Windows ML / DirectX layers
  • Multiple adapters/interfaces get listed as separate “GPUs”

Which one should you pick?​

Unfortunately, since they’re unlabeled, the only reliable way is a quick test:

Quick test method​

  1. Select the first RTX 3060 entry
  2. Trigger AI detections (walk in front of a camera, etc.)
  3. Open Task Manager → Performance → GPU
Look at:
  • “Compute” or “CUDA” usage spikes → best option
  • Minimal GPU usage + higher CPU → not ideal
Then repeat with the second entry.

✅ What you’ll usually find​

  • One option will:
    • Use more GPU
    • Use less CPU
    • Process alerts faster
That’s the one you keep.

⚠️ Bottom line​

  • You do not have two GPUs
  • This is normal for Blue Iris v6
  • It’s just two different ways to use the same RTX 3060
 
  • Like
Reactions: Skinny1