It's all about delay. One moment dashboard loads in a flash. Then 30 seconds. Motion detectors and lights of different protocols and manufacturers. Sometimes instantaneous. Sometime 10 seconds.
CPU often at 20 percent which is gone. But sometimes 70 without backups or anything special running. Memory never an issue. Drive space not an issue.
Logs show nothing. Disabled frigate cause I thought that might be it. No difference. Strong Omada network. Modern coordinators, Zooz and SLZB for z2m. Top notch hardware. Using HAOS 2026.6. Problems probably started about 6 months ago but it was insidious so I don't really know the trigger. I just can't detect a pattern.
I bought a new N100 with 26gb and 500gb, identical to my HA box. I'm thinking of installing HA and restoring backup.
I would suggest interference (neighbor with new wifi perhaps?) but since you mention both Zigbee and Z-wave I have no clue. Have you tried enabling debug logging to see if anything else pops up? Tried changing out the network cable to the NUC? Next time you get a stall, try pinging HA from another device, see if you have packet loss.
That pretty much describes my - very heavily loaded - current ha rig. (nuc, 250g storage 16g ram) I’d venture your problem isn’t the machine. (I run my LLM core ona different box)
Im with fles. Network. And unfortunately you’ll have to troubleshoot each independently and no it’s not a simple easy solution I'm afraid. But I would NOT move hardware based on what you describe.
Have you tried booting into safe mode to see if it's a custom integration messing with your system? I've had similar issues since April with Core consuming CPU and it started getting to the point of HA being almost unusable, yesterday I booted into safe mode and everything seemed to stabilise. Then just disabled all custom integrations and rebooted, still good. Now I'm just enabling them one at a time and keeping an eye on my CPU usage until something goes wrong again. It's a process but at least HA is stable again.
CPU is just only part of your indication of load. You'll need to monitor your 1m,5m, 15m load. If those are ' high' for you number of cores you have IO wait times which can come from anywhere
Sor for example, my 6 core cpu not breaking sweat at 12 -20% load, still had bad times with 4.5 1m load. Any event from/to ZHA or zwave just slowed down.
