How it works

Mining that doubles as real AI inference.

Pearl's Proof-of-Useful-Work means every "hash" is a matrix multiplication — the exact math an LLM runs. We turn that into a second income stream.

You point a miner at the pool

Datacenter H100/H200 connect through our inference gateway; consumer cards (3090/4090/5090) connect via standard pearl/v1 stratum. Two minutes, one command.

Your GPU computes NoisyGEMM

The Proof-of-Work IS a noised matrix multiply — identical to an LLM's forward pass. Your card earns PRL by doing math that's actually useful.

On the inference tier, that math serves real prompts

For H100/H200, the same compute that mines also answers paid inference requests — so the GPU-hour earns coins and inference revenue. (Proven end-to-end on live hardware.)

The pool accounts every share & pays you (PPLNS)

Shares are recorded in a crash-proof, idempotent ledger — no double-pays, ever. Payouts go out on the standard PPLNS schedule to your address.

One GPU-hour. Coins from mining + a share of paid inference.
That second stream is the revenue no fee-only pool can match.

FAQ

What does it cost? What's the fee?
1.5% all-in pool fee — 1.0% base + 0.5% on-chain tx reserve. PPLNS payout, 1 PRL minimum payout. The dashboard shows the live pool fee straight from the pool API — nothing hidden.
Which GPUs are supported?
The combined mine + inference tier — one gateway that switches between mining PRL and serving paid inference on the same card — needs the useful-work kernel, which is compiled for NVIDIA's sm90 architecture only: H100/H200. Consumer NVIDIA cards mine the pearlhash algorithm with PeakMiner (--coin pearl) over standard stratum — the one third-party miner we have verified supports the algorithm — and can separately register as a GPU serving node for inference, serving whatever model fits their VRAM (or a bigger one across a multi-GPU rig). SRBMiner-MULTI and BzMiner do not support pearlhash; we listed them in error and have said so on the miner compatibility page. AMD via Kryptex krig-miner needs a TLS stratum endpoint we do not yet offer. See the live hashrate table on the home page.
How and when do I get paid?
PPLNS — you're credited per accepted share, weighted across the payout window, and paid to your PRL address once you pass the minimum-payout threshold. Every credit and payout is in the durable ledger.
Is my balance safe? What about double-pays?
The pool's money ledger is an append-only, atomic SQLite store; payouts are idempotent on txid — a crash or retry can never pay you twice or lose a balance. An independent security audit gates real-money launch.
What is the "inference tier"?
On H100/H200, the NoisyGEMM that secures the chain is also a real LLM forward pass. We can route that compute to paid inference — but note: the "two income streams from one GPU" framing is not supported by measurement. An H100 serving inference produces ~0.67 MH/s of mining (1.8×10⁻¹⁴ of network hashrate) — effectively zero. Inference and mining are separate workloads on separate hardware. The pool supports both, but a GPU earns from one or the other, not both at once.
What does inference actually cost?
Measured 2026-08-10 on 1×H100 NVL ($2.60/h): Qwen3-30B-A3B at concurrency 256 sustained output at $0.107 per 1M tokens (6,747 tok/s aggregate, 281ms TTFT p95, zero errors). This is competitive with Together/DeepInfra for the same model class. The load curve was still rising at c=256, so $0.107 is an upper bound.
How do I connect?
See Connect in 2 minutes — copy the snippet for your tier (gateway for H100/H200, stratum URL for consumer cards), drop in your address, and you're earning.
Connect a miner →