Nvidia RTX (model not stated) · n/s · 40 cameras

Frigate 0.18.0-77a66e7 (tensorrt image)Community buildNew

Posted Oct 10, 2026·GitHub #24616 ·db-174

I run this too, add my numbers Opens the submit form prefilled with this hardware, model and version

Inference

17–22 ms

as reported, 4 detectors

Est. capacity

~205 det/s

1000 ÷ 19.5 ms × 4 · estimate

Cameras

as reported

CPU

—

not reported

Power

—

not reported

Parts

prices show only once verified by hand

Accelerator

Nvidia RTX (model not stated)

ONNX (Nvidia / GPU) · 1 build on DetectFPS

price not tracked yet

Get price drops in the weekly newsletter

Price drops, once a week

When a tracked price is verified or drops, it goes in the newsletter alongside new benchmarks. No per-device alerts and no tracking: the newsletter is the only list.

Newsletter coming soon The signup form appears here as soon as the newsletter embed code is pasted into site.json. Issue #1 is already readable here.

Host

—

not reported

—

CPU

AMD Threadripper 2990WX (VM, 48 vCPU)

as reported

—

Cameras

40 cameras

mixed 4K/1440p/1080p/D1 @5fps

—

Tracked totalno verified prices yet

Setup

Detector

ONNX (Nvidia / GPU)

Detector instances

Model, as reported

n/s

Model family · input

unspecified · input not stated

Camera resolution

mixed 4K/1440p/1080p/D1 @5fps

Cameras

Frigate version

0.18.0-77a66e7 (tensorrt image) (0.18 series, newest)

Ubuntu 24.04 VM (k3s)

Numbers

dashes mean the source didn't say

Inference

—

Inference range

17–22 ms

Total fps

—

CPU

—

Power

—

Used in stats

Yes

Notes and flags

ONNX detector on the 0.18.0 -tensorrt image (TensorRT execution provider), 4 instances at 17-22 ms; ~200 aggregate detect fps. Detectors kept up; the thread is a bug report about a frame-copy bottleneck elsewhere in Frigate.

Source

Where

github.com/blakeblackshear/frigate/discussions/24616

Date, as the source wrote it

2026-10-10 (day precision)

Added to DetectFPS

from a public post; numbers copied as written, not re-tested

What this build implies

40 cameras at 5 fps, Typical activity

Comfortable 100 of ~205 det/s

0%60%100%

Demand 40 × 5 × 0.5 = 100 det/s against ~205 det/s from this build's own 17–22 ms × 4 detectors. An estimate; the poster's note: “ONNX detector on the 0.18.0 -tensorrt image (TensorRT execution provider), 4 instances at 17-22 ms; ~200 aggregate detect fps. Detectors kept up; the thread is a bug report about a frame-copy bottleneck elsewhere in Frigate.”.

Try other settings in the calculator

Hardware

Nvidia RTX (model not stated)

Detector type

ONNX (Nvidia / GPU)

Builds on DetectFPS

Median, this model

19.5 ms (1 build)

Price

price not tracked yet

Affiliate links

DetectFPS may earn a commission when you buy through links on this site. It never changes which builds we show or how they rank. Buy buttons say “affiliate link” whenever one is used; none is used on this page.

Similar builds