FocsleFocsle
UL

YOLO27-Edge

18%
by Ultralytics

Ninth-generation YOLO with edge-first quantization recipe. The default workhorse for 1080p multi-stream perception.

Object detectionMITINT8INT4FP16cocoedgereal-time
487K downloads 38K deploymentsUpdated Apr 11, 2028
Headline:14.2ms · Hailo-8 · INT8

Cross-chip benchmark matrix

Every supported chip, in matched-pair runs from the Fo’c’sle HIL lab. Sortable by any column — click a header. Cells where the chip can’t run this model show Not supported.

Chip platformQuantLatency p50(ms)Latency p95(ms)ThroughputAcc. retention(%)Power(W)Memory(MB)Tested by
N
NVIDIA Jetson Thor
2070 TOPS · Module
FP166.277.76160 FPS99.6122.255Fo’c’sle HIL
N
NVIDIA Jetson AGX Orin
275 TOPS · Module
FP169.7711.6102 FPS99.449.455Community
Q
Snapdragon 8 Gen 3 NPU
45 TOPS · SoC
INT813.116.377 FPS98.06.328Fo’c’sle HIL
Q
Qualcomm QCS8550
48 TOPS · SoC
INT813.917.872 FPS97.913.028Fo’c’sle HIL
H
Hailo-10H
40 TOPS · M.2
INT817.022.659 FPS98.33.328Fo’c’sle HIL
N
NVIDIA Jetson Orin Nano
40 TOPS · SoM
INT818.122.055 FPS97.912.228Community
A
AMD Versal AI Edge VE2302
22 TOPS · SoC
INT818.123.955 FPS97.410.228Community
Apple Neural Engine (M4)
38 TOPS · SoC
FP1619.124.352 FPS99.97.155Community
A
Ambarella CV72
32 TOPS · SoC
INT820.725.448 FPS97.25.528Community
H
Hailo-8
26 TOPS · M.2
INT821.225.847 FPS97.52.328Fo’c’sle HIL
π
Raspberry Pi 5 + Hailo HAT
26 TOPS · HAT
INT825.934.039 FPS98.67.428Publisher
Q
Qualcomm QCS6490
12 TOPS · SoC
Not supported
A
Ambarella CV5
16 TOPS · SoC
Not supported
M
MediaTek Genio 700
4 TOPS · SoC
Not supported
G
Google Coral Edge TPU
4 TOPS · USB
Not supported
I
Intel Movidius Myriad X
4 TOPS · SoC
Not supported
R
Rockchip RK3588
6 TOPS · SoC
Not supported
Leader per columnCOCO 2017 val · 1080p input · single-stream, batch=1

HIL conditions

All numbers measured on Fo’c’sle HIL rigs in Tel Aviv (primary), Munich (secondary), and Pittsburgh (robotics). Single-stream, batch-1, real preprocessing, real downstream consumer. p50/p95 are over 10,000-frame steady-state windows after a 30-second warm-up. Power draw is package power, not wall power. Memory footprint is the resident model + activations footprint at peak — not on-disk.

Submitted publisher numbers are accepted only if they reproduce within ±8% of an HIL-lab matched run on the same chip in the same input mode. Otherwise they live separately under the Discussion tab.