I’m wondering if anyone else has suggestions and/or metrics around the configuration (CPU & Memory) of the VM they use to host HAOS?
The most taxing tasks I run seem to be Frigate (with CPU based detection) and periodic ESPHome compiles. The compiles are where I can run into issues.
As a benchmark, I’m using the ESPHome compile of wake-work-voice-assistant for an ESP32-S3-Box, after a full restart & clean build files. This is on a Ryzen 5800X host using Hyper-V. Results:
2 CPU, 4 GB, Frigate stopped - 658 seconds
2 CPU, 4 GB, Frigate running - 1272 seconds
3 CPU, 4 GB, Frigate stopped - 388 seconds
3 CPU, 4 GB, Frigate running - failed, Out-of-memory
6 CPU, 6 GB, Frigate stopped - 234 seconds
6 CPU, 6 GB, Frigate running - 301 seconds
Time roughly linear with available (not in use by Frigate) CPU count - CPU usage is pegged at 100% during compile.
Each extra CPU needs roughly 1GB extra memory.
Are other seeing similar (or conflicting!) numbers?
You should setup OpenVINO Object Detectors | Frigate
OpenVINO - thanks, yes, I’m running that, but still CPU based.
1 Like