Hardware
Can my machine run Dendra?
Give your graphics card, its video memory and your system memory. The page answers with the role your machine would have, the size of model it can run, and what the Final Testnet Season pays for that work.
The answer follows the kit's own hardware probe, deploy/hw_probe.sh, whose tiers are copied into a file this page reads. What you type stays in your browser.
Your machine
The kit uses NVIDIA cards only: its probe lists cards with nvidia-smi. Another make counts as no card.
Only shown back to you in the result.
With several cards, check each card on its own: bash deploy/hw_probe.sh --list-gpus prints a tier per card.
Leave it empty if you do not know: the page then says what it cannot decide, and does not count it as zero.
result
Answer the first question
Start with whether the machine has an NVIDIA graphics card. The other fields follow from it.
What the season can pay
reading the season status
The Final Testnet Season pays in testnet DNDR for work the chain records. The rates below are read from the season service now.
Each figure is a rate per unit of work and predicts no income. The programme answers a fixed number of requests a day for all present miners together, and the chain hands them out, so your share depends on how many identities are present. The rates do not depend on the tier of your card: pay follows counts of requests, windows and verdicts, and a judge day (final_season_calc.py::gross_of).
| Reward | Rate | How it is earned | For your machine |
|---|---|---|---|
| Verified request | ... | Per verified programme request the machine answers. | ... |
| Presence | ... | Per availability window proven on chain, up to a daily maximum. Paid only on a day the identity also served at least one verified request. | ... |
| Juror verdict | ... | Per verdict consistent with the outcome of an audit. A verdict needs at least audit_min_quorum jurors voting, and the jury excludes the miner under audit. Only a machine running a judge posts verdicts. | ... |
| Active judge | ... | Once a day to an identity that was an active judge that day, when it meets the conditions listed on the season page. | ... |
...
cap for the whole programme, per day (cap_programme_day)
...
cap for the whole programme, over the season (cap_season)
...
requests answered a day, all miners together (requests_per_day)
The tiers
The probe sizes the model to the total video memory of the card. It measures capacity and does not reserve it, so a card already used by a display or another model may fail to load the model.
...
| Tier | Card memory at least | Model budget at least | Models the kit may pick |
|---|---|---|---|
| reading the tiers file | |||
Judging
...
Read your real numbers
A box says 8 GB, 12 GB or 32 GB. The probe reads the exact total, and a tier can sit a few percent away from the figure on the box.
$ nvidia-smi --query-gpu=name,memory.total --format=csv
$ free -m
Under Windows, run it inside Ubuntu: the kit counts the memory of WSL 2, about half of the PC's by default. The memory= line of the [wsl2] section of %UserProfile%\.wslconfig raises it.
$ git clone https://github.com/DendraNetwork/dendra-network.git && cd dendra-network $ bash deploy/hw_probe.sh $ bash deploy/hw_probe.sh --list-gpus
Read the -> role line. The last command prints one line per card with its tier, for a machine with several. This page and the probe use the same tiers, and the probe is the one that decides at install time.
Run a miner
Install the kit, check it, and read what each role needs.
Open the miner guide
Final Testnet Season
The rules, the caps and how to check a payment.
Open the season page
Run a node
No GPU needed. The node and validator guide.
Open the node guide
Big hardware waitlist
A large card, several cards or a rig? Tell us what you run. Hardware only, no name or e-mail.
Open the waitlist