AMD XDNA2 NPU (Strix Halo)
AcceleratorAMD XDNA2 NPU
9 builds on DetectFPS: 9 Frigate docs references and 0 community builds. All 9 are usable for stats. Hardware id amd-xdna2-npu.
Median inference
20 ms
all models · 9 usable builds
Fastest model
7.4 ms
yolov9-t · 320 · 1 docs reference
Most cameras seen
—
no camera counts reported
Price
not tracked
no verified price yet
Max detectors
—
no limit recorded
Median inference by model
Capacity and cameras are estimatesModel · inputMedianCapacityCameras
yolov9-t · 320
7.4 ms
1 docs reference
135 det/s
32
yolov9-s · 320
9 ms
1 docs reference
111 det/s
26
yolov9-m · 320
13.1 ms
1 docs reference
76 det/s
18
yolov9-c · 320
14.1 ms
1 docs reference
71 det/s
17
yolov9-s · 640
20 ms
1 docs reference
50 det/s
12
yolov9-m · 640
34.4 ms
1 docs reference
29 det/s
6
yolov9-c · 640
35.2 ms
1 docs reference
28 det/s
6
yolov9-e · 320
69.4 ms
1 docs reference
14 det/s
3
yolov9-e · 640
224.8 ms
1 docs reference
4 det/s
1
Bars show relative throughput (fastest model = full). Capacity = 1000 ÷ median ms for one detector. Cameras = how many at 5 fps and Typical activity stay under 60% of that capacity. Where a row mixes docs and community numbers, treat the median as a range rather than a point.
All builds for this device
Open in the database →BuildSourceModelSpeedCamsFrigatePosted
Community builds first, then docs references.