Guide · Nvidia RTX 2060
Nvidia RTX 3050 for Frigate: YOLOv9 and RF-DETR Inference Times
RTX 3050 Frigate benchmarks with the ONNX detector, plus RTX 3060/3070/5060 Ti comparisons and driver requirements.
last reviewed 2026-10-19
newest Frigate 0.18.2Hardware in this guide: Nvidia RTX 2060 · Nvidia RTX 3050 · Nvidia RTX 3060 · Nvidia RTX 3070 · Nvidia RTX 5060 Ti · Nvidia RTX 5070 Ti · Nvidia RTX A4000
Nvidia GPUs run Frigate detection through the onnx detector (CUDA Graphs since 0.17). The RTX 3050 is the cheapest card in the docs table; we include other Nvidia rows for context.
Known issues and gotchas
- Requirements: host driver ≥545, Compute Capability ≥5.0, nvidia-container-runtime (docs). GTX 900-series dropped in 0.17 (0.17.0 notes).
- YOLO-NAS is not accelerated with CUDA Graphs (marked ❌ in docs); YOLOv9 and RF-DETR are (docs).
- Use the TensorRT execution provider: an RTX 3060 user went from 28–34 ms to 10–14 ms on YOLOv9-t by enabling it (db-140).
- Power: a dGPU draws far more than an iGPU/NPU; an RTX 2060 build reports 95–150 W loaded (db-132/133).
- CUDA Graph hang reported on an RTX 5070 Ti in 0.17.2 (db-155).
Related
Live numbers for this guide's hardware
Generated from the database at build timeThe text above is written by hand and never contains numbers or prices. Everything below is pulled from builds.json and hardware.json when the site is built, so it stays current without touching the guide.
Nvidia RTX 2060
AcceleratorTensorRT (Nvidia, legacy)
Median inference
30.6 ms
all models · 2 usable builds
Capacity · estimate
~57 det/s
yolov7x · 320 median 17.4 ms · 1 detector
Cameras · estimate
13
5 fps · Typical · under 60% (22 before overload)
Price
not tracked yet
no retailer link listed
BuildSourceModelSpeedCamsFrigatePosted
Get price drops in the weekly newsletter
Nvidia RTX 3050
AcceleratorONNX (Nvidia / GPU)
Median inference
11 ms
all models · 6 usable builds
Capacity · estimate
~125 det/s
yolov9-t · 320 median 8 ms · 1 detector
Cameras · estimate
30
5 fps · Typical · under 60% (50 before overload)
Price
not tracked yet
retailer link below
BuildSourceModelSpeedCamsFrigatePosted
Get price drops in the weekly newsletter
Nvidia RTX 3060
AcceleratorONNX (Nvidia / GPU)TensorRT (Nvidia, legacy)
Median inference
10 ms
all models · 3 usable builds
Capacity · estimate
~167 det/s
yolov7 · 320 median 6 ms · 1 detector
Cameras · estimate
39
5 fps · Typical · under 60% (66 before overload)
Price
not tracked yet
no retailer link listed
BuildSourceModelSpeedCamsFrigatePosted
Get price drops in the weekly newsletter
Nvidia RTX 3070
AcceleratorONNX (Nvidia / GPU)
Median inference
8.5 ms
all models · 6 usable builds
Capacity · estimate
~167 det/s
yolov9-t · 320 median 6 ms · 1 detector
Cameras · estimate
39
5 fps · Typical · under 60% (66 before overload)
Price
not tracked yet
no retailer link listed
BuildSourceModelSpeedCamsFrigatePosted
Get price drops in the weekly newsletter
Nvidia RTX 5060 Ti
AcceleratorONNX (Nvidia / GPU)
Median inference
6 ms
all models · 4 usable builds
Capacity · estimate
~250 det/s
rf-detr-nano · 320 median 4 ms · 1 detector
Cameras · estimate
60
5 fps · Typical · under 60% (100 before overload)
Price
not tracked yet
no retailer link listed
BuildSourceModelSpeedCamsFrigatePosted
Get price drops in the weekly newsletter
Nvidia RTX 5070 Ti
AcceleratorONNX (Nvidia / GPU)
Median inference
10.5 ms
all models · 1 usable build
Capacity · estimate
~95 det/s
yolov9-s · 320 median 10.5 ms · 1 detector
Cameras · estimate
22
5 fps · Typical · under 60% (38 before overload)
Price
not tracked yet
no retailer link listed
BuildSourceModelSpeedCamsFrigatePosted
Get price drops in the weekly newsletter
Nvidia RTX A4000
AcceleratorONNX (Nvidia / GPU)
Median inference
15 ms
all models · 1 usable build
Capacity · estimate
~67 det/s
yolo-nas · 320 median 15 ms · 1 detector
Cameras · estimate
16
5 fps · Typical · under 60% (26 before overload)
Price
not tracked yet
no retailer link listed
BuildSourceModelSpeedCamsFrigatePosted