AcuPick

Get it from Andy...

608-XI have big drop in network bandwidth/decoding capabilities when you enable AI on NVR...
You have 12 Mpx cams, so for you this is a big problem :)

new (2025) NVR's (like 5xxx-EI2 or not released yet 5xxx-XI-PRO with WizSeek) based on never chipsets don't have this problem (difference between AI on/off is very small)...
 
steve1225 said:
Get it from Andy...

608-XI have big drop in network bandwidth/decoding capabilities when you enable AI on NVR...
You have 12 Mpx cams, so for you this is a big problem :)

new (2025) NVR's (like 5xxx-EI2 or not released yet 5xxx-XI-PRO with WizSeek) based on never chipsets don't have this problem (difference between AI on/off is very small)...
Click to expand...
These products are way overpriced compared to their performance. I’m kind of disappointed with my investment in this NVR. I think the firmware isn’t capable of handling all the AI features listed in the menus it feels half cooked. What I don’t understand is, when I turn on the AI features on the NVR side, the CPU stays below 10%, memory stays under 30%, and bandwidth usage is below 100 Mbps, yet the recordings and live streams still drop frames. I just don’t get it.
 
sunny0101786 said:
These products are way overpriced compared to their performance. I’m kind of disappointed with my investment in this NVR. I think the firmware isn’t capable of handling all the AI features listed in the menus it feels half cooked. What I don’t understand is, when I turn on the AI features on the NVR side, the CPU stays below 10%, memory stays under 30%, and bandwidth usage is below 100 Mbps, yet the recordings and live streams still drop frames. I just don’t get it.
Click to expand...

NVR is hardware products (not software on Windows) - it have very hard spec sheet and max limits. They guarantee how many stream can be decoded with/without AI and don't allow to more even if NVR is doing nothing.

You have specific configuration - You use 12Mpx cameras - which creates a lots of problems (we warned You about them :)...
 
  • Like
Reactions: looney2ns and sunny0101786
ps. 5xxx-EI2 (much cheaper than 608-XI) have ability to decode 6 cameras 12Mpx@30fps without AI and 5 cameras 12Mpx@30fps with AI...
It's 2025 (not 2022/2023) NVR and has next generation of chipset inside.. You can sell 608-XI and replace it with 5xxx-EI2...\

it's first Dahua NVR model which can display/decode 20 channels at 4Mpx@30fps without AI and 16 channels at 4Mpx@30fps with AI..

ps. to BI users - how many 4Mpx@30fps channels can BI decode/display with modern GPUs???


Screenshot 2025-09-16 at 15.49.01.png


ProColor Series Camera

Building security systems for home and business offices. Protecting customers' personal safety and property security. All the security cameras and NVRs we sell are of great quality and at affordable prices.
empiretech01.com empiretech01.com

All Andy AI-S2 NVRs models are from 5xxx-EI2 line...
 
Last edited:
  • Like
Reactions: Jahua95
steve1225 said:
ps. 5xxx-EI2 (much cheaper than 608-XI) have ability to decode 6 cameras 12Mpx@30fps without AI and 5 cameras 12Mpx@30fps with AI...
It's 2025 (not 2022/2023) NVR and has next generation of chipset inside.. You can sell 608-XI and replace it with 5xxx-EI2...\

it's first Dahua NVR model which can display/decode 20 channels at 4Mpx@30fps without AI and 16 channels at 4Mpx@30fps with AI..

ps. to BI users - how many 4Mpx@30fps channels can BI decode/display with modern GPUs???


View attachment 228197


ProColor Series Camera

Building security systems for home and business offices. Protecting customers' personal safety and property security. All the security cameras and NVRs we sell are of great quality and at affordable prices.
empiretech01.com empiretech01.com

All Andy AI-S2 NVRs models are from 5xxx-EI2 line...
Click to expand...

You know most BI users don't run over 15FPS lol.....but an 8th gen without a GPU could handle 12 4MP cameras running at 30FPS mainstream only (and a lot more if substreams are used). I don't have a modern GPU to test how much more it could handle LOL.
 
  • Like
Reactions: sunny0101786
Chat GPT answered to my NVR issue:
it looks like your NVR has plenty of CPU, memory, and network headroom, but frames are still dropping when AI features are enabled. That usually means the bottleneck isn’t in the places you’re checking. Here are some possible reasons:
Likely Causes of Frame Drops
  1. Dedicated AI Hardware Limits
    • Many NVRs have a separate chip or DSP for AI tasks (e.g., face detection, license plate recognition). Even if CPU and RAM are low, that AI processor may be maxed out, causing dropped frames.
  2. Firmware / Software Inefficiency
    • The firmware may not be optimized—AI processing could interfere with video decoding/encoding pipelines, introducing lag or frame skipping even when resources look fine.
  3. Codec/Resolution Processing Strain
    • High-resolution streams (8MP/12MP) require a lot of decoding before AI analysis. The NVR may handle storage fine, but AI analysis adds overhead per stream that isn’t reflected in CPU usage.
  4. Disk I/O Bottleneck
    • If the drives can’t keep up with simultaneous recording and AI indexing, the NVR may prioritize storage, dropping frames in live view or recordings. (This is definitely not the case because I tested on RAID 0 x 4 drives)
  5. Internal Bus / GPU Limitations
    • The bottleneck could be the data pipeline between the decoder, AI chip, and encoder—not CPU/RAM. Frames might be dropped because the bus can’t move data fast enough.
  6. Firmware Bugs or Incomplete AI Implementation
    • Since you mentioned it feels “half-cooked,” it’s possible the AI functions aren’t stable and cause desynchronization in the recording pipeline.
  7. Camera-NVR Sync Issues
    • When AI is enabled, the NVR might request additional metadata from the cameras. If the sync isn’t handled well, you get frame drops.
  8. Insufficient GPU / NPU Resources (hidden from system stats)
    • System stats (CPU/memory) don’t always show utilization of hardware accelerators like GPUs or NPUs. These might be maxed out.
In short: the “hidden bottleneck” is likely in the AI chip / decoding pipeline, not in CPU, RAM, or network.
 
steve1225 said:
which AcuPick function?

search by image (clicking on icon next to each person/vehicle on footage to find any other footage of similar looking persons/vehicles) or search by attributes (show me all males in white t-shirt or blue suv from toyota)?
Click to expand...
Search by image - people
 
You must log in or register to reply here.
Top Bottom