Lab dashboard
Miner performance lab · zkPoW & matmul PoUW

Miner Performance Lab

Try GPU + miner settings and see predicted proof-rate and profitability without renting or running a thing. The current zkPoW regime is calibrated on real goldenminer measurements we collected; the Fork A “matmul PoUW”regime models the upcoming change where mining becomes matrix-multiply AI work — so you can see how your hardware's value shifts before it lands, and whether tensor-heavy cards (or datacenters merge-mining AI) are about to out-earn you.

Calibrated estimate — not a guarantee

The current zkPoW regime is calibrated on 7 real goldenminer measurements (fit R²= 0.989, ±4.6 p/s). The Fork A “matmul PoUW” regime is now calibrated on 3 real GEMM-throughput benchmarks we ran — measured matmul is ~half the spec-sheet peak, and consumer/workstation cards run tensor at half rate (real/spec ≈ 0.29) while full-rate datacenter parts hit ≈0.71. Cards we didn't measure use their architecture's measured factor (wider band). It still models a not-yet-live protocol — treat it as a forecast, not a promise.

Hardware & settings

GPU (incl. cards you don't own — that's the point)

PoW regime

Miner mode (--mode)

threads-per-card: 4

GPU count: 1

Economics

NOCK price (USD)

rent $/hr per card

Predicted (calibrated)

proof-rate (1× RTX 4090)135 p/s

± 5 p/s · 135 p/s/card before settings

proof-rate per $ rent/day

14.1 p/s/$

NOCK / day

6.77

share 6.8e-3%

revenue / day

$3.39

rent / day

$9.60

profit / day

-$6.21

NOCK/day depends on your network-size & emission snapshot (defaults: 2,000,000 p/s network, 100,000 NOCK/day). The robust, calibrated outputs are proof-rate and per-$ value.

How RTX 4090's rank shifts when matmul PoUW lands

Among 24 cards, the RTX 4090 ranks #3 under today's zkPoW but #5 under Fork A matmul PoUW — down 2 (FP32 advantage fades; datacenter tensor silicon overtakes it). Datacenters merge-mining AI work at ~zero marginal cost is the dilution this models.

Catalog ranked by predicted p/scurrent zkPoW (calibrated) · gpu · 4 thr · 1×

dark = selected · gray = datacenter card · light = consumer · click a bar to select

Honest limits

Current-regime predictions are a least-squares fit over a handful of real single-card runs — good for ranking and rough sizing, not a guarantee for your exact host. Fork A is a modeled estimate: the matmul PoUW protocol change isn't live, so those numbers track spec-sheet tensor throughput and a placeholder dilution/emission snapshot you can override. Real matmul-PoUW calibration data gets folded in as the change ships.