GPU Watchdog
reading device…
live
▲
—
Stop request
Free the GPU
Force stop
Resident in VRAM
nothing loaded
Recent model swaps
none
Temperature
—
°C
slowdown — · shutdown —
Utilisation
—
%
SM clock —
VRAM
—
MiB
of — MiB
Power
—
W
no model resident
Thermal & load
last 30 min · live
temp °C
util %
throttled
In flight
Usage by app
attributed from container source IP
1H
6H
24H
7D
Apps
App
Reqs
Out tok
Avg
p95
tok/s
GPU s
no requests recorded yet
Models
Model
Reqs
Out tok
tok/s
Cold starts
no requests recorded yet
Recent requests
newest first
When
App
Model
Endpoint
Duration
In
Out
tok/s
Temp
waiting for the first request through the proxy…