Your GPUs
Loading…
Know the moment a rented GPU stops delivering.
Hosts lower power limits mid-rental, cards throttle when a room heats up, links renegotiate down. A one-time check can't see any of it. Monitor watches every GPU while your work runs, emails you when something changes, and keeps the history.
One command per node
Our open-source checker, tokenwatt-check, samples every GPU once a minute and reports every five. No agent to trust blindly: read the code.
Alerts that mean something
Power limit cut, hardware or thermal throttling, clock collapse under load, PCIe link drop, new memory errors, a node that stops reporting. Measured against what that GPU model should do.
History you can show a provider
Every reading is kept: when the power limit changed, how long a card throttled. Evidence for a refund or a better contract, not a hunch.
What runs on your node
pip install -U tokenwatt-check
export TOKENWATT_KEY=twk_… # from your sign-up email
nohup tokenwatt-check --watch --monitor --node-name "$(hostname)" &
It sends only GPU telemetry: power limit and draw, SM clock, temperature, utilization, PCIe link and ECC error counts, plus the events it detected. No hostnames unless you name the node, no processes, files or environment variables. The key is read from an environment variable or a file, never the command line.
Start monitoring
Free during early access, up to 64 GPUs, NVIDIA for now. We email you a key for your nodes and a private link to your dashboard.