What a card mines, by tier.

The schedule the figures use

EmissionSchedule::TESTNET_1 (the testnet genesis, decided 7 October 2026): 100 IGN a block at one block a second, a 90-day ramp from 10 percent, a monthly glide with a two-year half-life (each month pays 2^(-1/24) of the month before), then 1 percent of supply a year from about year 11.4. Of each block, 80 percent goes to the miner who found it and 20 percent to the proving pool. A miner whose key signs its checkpoints keeps the full 80 percent; an unsigned key gives a tenth of it to the proving pool.

A solo card's expected income is its share of network hash times 86,400 blocks a day times the figure above. The tables use day 90, when the ramp is over. Before that, multiply by the ramp row; after that, by the glide row. A pool user gets the same expected amount minus the pool's fee (pool-0: 1 percent) with the variance taken out; the reference pool holds no balance and pays from the coinbase split.

The cards, measured

6 October 2026, rented single-card Linux boxes (docs/analysis/prover-tiers-real-cards.md, bench log 'Prover tiers on real cards: the rented fleet, 6 October 2026'): the 0.3.12 miner on the ten-field override, the miner alone on the card, nvidia-smi power at the same second. The RX 9070 XT row is the telemetry run of 5 October 2026 on a Windows rig (bench log 'Power, heat, fans and clocks, measured'); the Apple row is the Metal bench of 4 October 2026 (site/miner-bench.json, class v4). The class v4 rates on the live devnet run lower than the ten-field rates on the same card (RTX 5090: 124.2 MH/s class v4 against 98.5 on the rented box under a different driver and container, docs/bench-log.md); the table is re-generated at the testnet cut from that class.

IGN a day per card, solo, after the ramp (day 90)

Three network sizes, because income is a share of the network and nobody knows the network before it exists. 1 GH/s, 10 GH/s, 100 GH/s are reference sizes, not forecasts. The live network's estimate is on /live; divide by it.

Electricity a day, and the electricity cost of one mined IGN

Electricity a day = measured watts x 24 h x the tariff. The cost of one IGN divides that by the day-90 figure at 10 GH/s; at another network size it scales with the network (ten times the network, ten times the cost per coin). The three tariffs: USD 0.05 a kWh is cheap industrial power, USD 0.10 is a US retail rate, USD 0.25 is a UK retail rate at today's level (each approximate; tariffs move, the watt figures do not).

What it means per tier, and what is being done

• One 8 GB card. NVIDIA RTX 4060 Ti 8 GB: 19.1 MH/s at 73 W, 12,441 IGN a day at 10 GH/s after the ramp, electricity 0.174 a day at USD 0.10 a kWh. Mines; proves only core-only shards (docs/analysis/prover-tiers-real-cards.md). The record says this tier is first under power in every exit (docs/analysis/mission/past.md section 3): the glide never halves it overnight, and the card games when the income goes. Being done: the Cards screen shows the expected MH/s, W and blocks a day before the first share (mission item 6).
• One 12 GB card. NVIDIA RTX 5070: 41.9 MH/s at 137 W, 27,329 IGN a day at 10 GH/s after the ramp, electricity 0.329 a day at USD 0.10 a kWh. Mines and proves beside the miner on the patched server. Being done: the same Cards line; the proving tier sentence names the shard size the card takes.
• One 16 GB card. NVIDIA RTX 4060 Ti 16 GB: 17.6 MH/s at 72 W, 11,469 IGN a day at 10 GH/s after the ramp, electricity 0.174 a day at USD 0.10 a kWh. Mines and proves; the AMD row is the measured 9070 XT, 199 W for 18.6 MH/s, so per watt it is about a fifth of the NVIDIA cards of its tier and the electricity cost per coin is the highest in the table. Being done: the AMD read-width experiment (docs/plans/read-width.md); until it lands the AMD tier is told its per-watt figure before it buys.
• One 24 or 32 GB card. NVIDIA RTX 5090: 98.5 MH/s at 258 W, 64,249 IGN a day at 10 GH/s after the ramp, electricity 0.620 a day at USD 0.10 a kWh. The last GPU standing in every exit on record, and the only tier that proves the full shard beside the miner. Being done: the proving pool is the second income for this tier and the Earnings screen shows shards and IGN beside blocks.
• A rig. 6 x NVIDIA RTX 4070: cards times the card figure; the host's own draw is not measured and is owed. Rigs followed income across chains inside days on every chain in the record; Igneum expects no loyalty and pays staying keys a vote (30 days of blocks).
• A pool user. The same expected IGN minus the pool's fee, variance removed; pool-0 holds no balance and the member's own key is in every block it finds, so a pool's vote is the member's weight. Being done: the pool finished (mission item 11).
• Windows, Linux, macOS. The rates above are Linux containers (NVIDIA) and Windows (AMD); the Mac row is Metal on Apple silicon with no power reading. The same card on another operating system is within the driver's margin, not measured here: owed per platform at the testnet cut.