Hi ​@fatima-zahra.chenani! Great to see you here! Ah interesting, yes. This is expected, yeah, let’s look at why it’s probably happening:

• Higher latency with 4 cores: Using multiple AIPU cores adds synchronisation overhead. For single-image inference, this could actually increase latency slightly.

• No FPS gain: If your pipeline isn’t bottlenecked by AIPU compute time (e.g. single stream), then extra cores don’t boost throughput.

• Increased CPU usage: Managing more AIPU cores increases host-side coordination, so CPU load naturally goes up a little.

Considering these, inference tool guide and deployment options, we could look at it in terms of:

• Use 1 core for low-latency applications.

• Use 4 cores for high-throughput, multi-stream workloads.

Let me know if that helps, and we can dig deeper!

Hi, which host is it you are using, and what is your source, is it a usb or rtsp fixed frame rate source.

As Spanner says we do currently expect higher latency with more cores involved. But also we get better throughput but if your source is frame limited then you wouldn’t see that.  And I would not expect the cpu usage to go up like that.

Could you try with a video file and tell us what the throughput is.