Start here: are you sure anything is wrong?
Solo mining is lumpy by nature. You're paid the full block reward when your work finds a block, and nothing in between. If your share of total network compute is small, going days without a block is statistically normal, not evidence of a fault.
So before troubleshooting, work out roughly what to expect: your share of network compute × blocks per day (about 445 at Pearl's 194-second block time). If that comes to a fraction of a block per day, long dry spells are simply the maths.
That said, here's how to confirm your setup is genuinely working.
Check the stack is running
Pearl mining needs three things alive at once:
pearld— your node, synced and connected to peerspearl-gateway— connected to the node over JSON-RPCvllm-miner— serving the model on your GPUs
If any one is down or misconfigured, you're producing nothing. Check each in turn rather than assuming.
Work through these
1. Is your node fully synced?
A node still catching up isn't building on the current tip, so any work you do is wasted. Check your node's block height against the current height on explorer.pearlresearch.ai. They should match.
2. Are you on the right network?
A node started with --testnet or --simnet is mining a test network with valueless coins. Easy to leave set after testing. Confirm you're on mainnet.
3. Is your mining address set correctly?
Check --miningaddr on the node and PEARLD_MINING_ADDRESS for the gateway. It must be a valid Taproot address beginning prl1p.
Check the running configuration, not just your notes. A typo here means blocks you mine pay someone else — and that's unrecoverable.
4. Is the gateway actually connected to the node?
Confirm PEARLD_RPC_URL, PEARLD_RPC_USER, and PEARLD_RPC_PASSWORD match your node's settings, and that the port is right for your network — 44107 for mainnet RPC. The gateway exposes its mining interface on /tmp/pearlgw.sock, or TCP 8337 if MINER_RPC_TRANSPORT=tcp.
5. Are your GPUs actually being used?
Run nvidia-smi and confirm utilisation and memory use consistent with serving the model. Near-zero utilisation means the miner isn't doing work — check its logs for startup failures, most often CUDA version mismatches or insufficient VRAM.
6. Are you running a Pearl-certified model?
Mining requires a Pearl-certified model such as pearl-ai/Llama-3.3-70B-Instruct-pearl. A standard model has no mining instrumentation in its forward pass — it will serve inference perfectly and mine nothing.
7. Read the logs
Start the node with --debuglevel=debug and check the miner and gateway logs. Most problems announce themselves clearly — connection refused, model load failure, CUDA error, invalid address.
Prove the pipeline end to end
The cleanest way to separate "not configured" from "unlucky" is to run against simnet, where difficulty is trivial and you should find blocks almost immediately.
If simnet produces blocks, your stack works and mainnet is a matter of share and variance. If simnet produces nothing, you have a real configuration problem.
I found a block but didn't receive the reward
- Check the payout address on the explorer:
https://explorer.pearlresearch.ai/address/followed by your address. Rewards appear there, not in your miner's output. - Confirm it's the address you configured, and that it's a wallet you control.
- Check your wallet has synced — a display lag is common. See Why does my wallet show a 0 balance?
- Orphaned blocks occasionally happen: two miners find a block near-simultaneously and only one survives. Rare, and the explorer is the source of truth.
Rewards are lower than expected
Not a fault. Your share of rewards falls as network compute grows, and the block subsidy declines gradually by design. See How mining on Pearl works and Does Pearl have halvings?
And to be direct: no software or setting can increase your rate above what the protocol pays. Anything promising otherwise is a scam — see Third-party mining software: why we don't endorse it.
Getting help
Our Discord is the fastest route — other operators recognise these symptoms quickly. Bring:
- Your GPU setup and driver/CUDA versions
- Which network and which model
- Relevant log output — with credentials and addresses removed
- Whether simnet produces blocks
If you think it's a bug, see Reporting a non-security bug.