Docs · $GPU

The loop

Buy a graphics card with tokenised NVDA. Rack it. Pay its power bill in $GPU, which is burned. It mines $GPU until it wears out.

The loop in ten seconds.

Every part of that sentence moves value in one direction. The NVDA you spend on hardware goes to the treasury and the liquidity pool. The $GPU you spend on electricity is destroyed. Nothing in the game pays anyone out of thin air.

Cards

Four models. More compute means more watts, a bigger price and a longer life. Prices sit on a stepped bonding curve, so each purchase makes the next one dearer.

Model	Compute	Draw	Life
SPINDLE 40 · Common	180 TF	150 W	45 d
SPINDLE 90 · Rare	520 TF	350 W	75 d
SPINDLE 100 · Epic	1,400 TF	700 W	120 d
SPINDLE 200 · Legendary	3,600 TF	1,200 W	180 d

Cards do not die suddenly. As they wear, compute falls while the power draw stays flat — so a card slowly stops covering its own bill and switches itself off. That is the whole ageing mechanic: obsolescence, not a timer.

Racks & watts

A rack has eight bays and a watt ceiling. A card needs a free bay and spare watts. That double constraint is the puzzle — you cannot simply stack legendaries.

Five bays free, and the card still doesn't fit. Watts run out before space does.

A cooler takes a bay of its own and cuts the watts of every card in the rack. Giving up a bay to power the rest is the first real trade the game asks you to make.

Rooms

Racks live in a room, and the room has its own two ceilings: how many racks it holds and how much power its grid connection can deliver. Both rise together when you upgrade — space you cannot power would be a trap, not a choice.

Closet → server room → machine hall → data centre.

Room	Rack bays	Grid feed
Closet	2	4 kW
Server room	4	10 kW
Machine hall	8	24 kW
Data centre	16	60 kW

Emission

800,000,000 $GPU is set aside for mining and released across five epochs. Each epoch emits half as much as the last, at half the rate — so every epoch lasts the same length of time.

Five process nodes. Each wafer packs twice the dies of the one before — and pays half as much.

Epoch	Node	Emits	Rate
1	5 nm	400M	base
2	4 nm	200M	÷2
3	3 nm	100M	÷4
4	2 nm	50M	÷8
5	1.4 nm	25M	÷16

The halving is triggered by how much has been mined, never by a clock. If the network stalls, the epoch simply lasts longer.

Your share

Rewards are split by the square root of installed compute:

\text{share}_i \;=\; \frac{\sqrt{\text{tflops}_i}}{\sum_j \sqrt{\text{tflops}_j}}

Square root, not a straight ratio. Going from 1,000 to 4,000 TFLOPS only doubles your cut — large capital scales slowly, and the board stays open to new players.

There is no cooldown anywhere in the design. The cost of growing is always capital, never waiting, because a cooldown hands a permanent moat to whoever arrived first.

The power bill

Electricity is not a fixed price. It is set so that a constant share of everything mined gets burned:

\text{rate} \;=\; \frac{\alpha \cdot E_{\text{day}}}{\max\!\left(W_{\text{active}},\; W_{\text{floor}}\right)}

α = 0.55. The floor in the denominator matters: without it, the very first miner on an almost-empty network is handed a near-infinite bill.

\text{bill}_i \;=\; \text{watts}_i \cdot \text{rate} \cdot \text{days}

Multiply the bill by every card and the network total is always α × emission. So 55% of everything mined is destroyed, whatever the price does and however large the network gets. A cheap token does not mean cheap electricity.

Three consequences fall out of that single line:

— The rate drops as the network grows. Economies of scale, without coding them.
— It halves at every halving, alongside the reward, so margins survive the cliff.
— Reward follows √compute but the bill follows watts linearly, so efficiency — compute per watt — is the actual skill.

Which gives the only formula that decides whether a card is worth running:

\frac{\sqrt{\text{tflops}_i}}{\text{watts}_i} \;>\; \alpha \cdot \frac{\sum_j \sqrt{\text{tflops}_j}}{\sum_j \text{watts}_j}

Your efficiency against the network average. When weak cards switch off the average rises, so the bar rises with it.

Bills don't go to a treasury, a team wallet or a staking pool. There is no exit from the furnace.

Unpaid cards go dark

A card with an unpaid bill stops mining, stops wearing, and leaves the network totals entirely — so it does not dilute anyone else's rate while it sits idle.

The floor

Every card bought sends 55% of its price to the treasury in NVDA. Every power bill destroys $GPU. One side rises, the other shrinks:

\text{floor} \;=\; \frac{\text{treasury NVDA}}{\text{circulating \$GPU}}

Both sides are denominated in the same asset, so the number reads without any exchange-rate noise. No mechanism in the game pushes it down — only the price of NVDA itself moves it, and that is the same price as the collateral.

The numbers

Bucket	Share	$GPU
Mining rewards	80%	800,000,000
Pool liquidity	20%	200,000,000
Team, presale, airdrop, advisors	0%	0

Supply is fixed at 1,000,000,000 and can never rise. The liquidity share is deposited one-sided above the going price, so no NVDA is supplied at launch — it accumulates as people buy. There is nothing pre-mined to sell.

Live figures are on Stats.

What can go wrong

Our contracts are immutable and have no admin key. The thing backing them is not.

The collateral has an owner

Tokenised NVDA on Robinhood Chain sits behind an upgradeable beacon proxy shared by every stock token. The issuer can replace its logic, freeze all transfers, block a named address, and burn tokens straight out of a wallet. The floor depends on that collateral, so it depends on them. Full detail on Stats.

Tokenised NVDA is a debt security issued by Robinhood Assets (Jersey) Limited. It gives you economic exposure, not a claim on the underlying share. We found this in the preflight audit and publish it rather than bury it.

Beyond that: the contracts are not deployed yet, every figure on this site is simulated until they are, and a mining game with thin liquidity behind its collateral is a risk in its own right.