Fab

Mine NVIDIA with an NVIDIA. 60 seconds of GPU time, paid in tokenized NVDA, TSM and SK hynix. Robinhood Chain.

0xc7db9a7034d061feca50f0a25636bcc83ca717d9

[Fab @Fab_RWA]

Your GPU has three parents. NVIDIA designed it, TSMC etched it, SK hynix built the memory. You paid all of them once.

FAB is live: 60 seconds of GPU time, paid back in their stock tokens on Robinhood Chain.

CA: 0xc7db9a7034d061feca50f0a25636bcc83ca717d9

https://t.co/i8TvLGVqX5

[Fab @Fab_RWA]

How we count an inference token so it can be paid.

A forward pass over 128 tokens costs 22.3 GFLOP in this shape, 174 MFLOP per token. The fab counts what the GPU verified, not what the browser claims: GPU timestamp queries time the kernels, the score is capped at the ceiling for your vendor's silicon, and a run only counts if it comes back after the full 15 minutes. 40 TFLOP of verified work is one compute unit, one compute unit is $0.0009 of stock tokens.

The weights are synthetic, so the tokens mean nothing yet. The compute is real, which is the part that gets paid. Real weights and real prompts are the next step, and the accounting does not change.

[Fab @Fab_RWA]

Someone asked why FAB runs benchmarks and not real GPU jobs. Fair. Two new instruments shipped tonight.

FAB-05 INFERENCE runs a full transformer forward pass in your browser: 12 layers, 768 wide, 12 heads, 128 tokens, every op a real model runs. Layernorm, attention with softmax, GELU MLP, residuals. Measured on an M2 Pro: 2,178 tokens per second, sustained over the whole run.

FAB-03 RENDER is a path tracer: 1024², four bounces, 1.5 billion paths per second on the same laptop.

Same rules as before. Timed by the GPU's own clock, signed by the fab, paid in the stock tokens of whoever built your chip.
https://t.co/sjzCAm4SBD

[Fab @Fab_RWA]

Robinhood Chain has about 2,000 stock tokens and very few people holding them on chain. Nobody had a reason to.

FAB gives GPU owners one. Run the card, get paid in the shares of the companies that built it. NVDA, TSM, SK hynix, AAPL, delivered to your wallet by a contract, not a promise. The chain was built for tokenized stocks. FAB is the first thing on it that pays them out for work.

[Fab @Fab_RWA]

What $30 of $FAB gets you.

A mining pass: 15, 30 or 60 minute runs, as many as four an hour, paid in stock tokens. At today's rate a 4090 makes about $1.05 per 15 minutes, a 4070 about $0.55, an M2 Pro about $0.17.

The rate moves with the treasury, never with a promise.

https://t.co/sjzCAm4SBD

[Fab @Fab_RWA]

Whatever is in your machine pays you. A 4090 pays NVDA, TSM, SK hynix. A MacBook pays AAPL, TSM, SK hynix. A Radeon pays AMD and TSM. An Arc pays Intel. FAB reads the chip, not your wallet, and the split follows who built it.

Bring your hardware.

[Fab @Fab_RWA]

$FAB in one paragraph for people who read contracts.

1,000,000,000 supply, all of it launched on the Pons v2 curve. 0% to the team. Liquidity locks permanently at graduation, nobody can pull it, including us. No emissions and no staking: $FAB is a key, not a reward. The rewards are stock tokens from a treasury contract you can read, paid to people who run GPUs. The token gates the fab, the fab pays in shares, and the two never mix.

[Fab @Fab_RWA]

The first voucher FAB ever signed, as a picture. Run 84004, Apple M1, 15 minutes of work, $0.11, four stock tokens, one transaction to claim. Small on purpose. The point is that it exists on chain and anyone can check it.

[Fab @Fab_RWA]

This is what a FAB payout looks like on chain.

Epoch 82810 closed in block 56321677. One Apple M1 in France ran for its trial, earned $0.11, and the fab signed it into 0.000158 AAPL, 0.000082 TSM, 0.000075 SK hynix and a sliver of the semis ETF.

The voucher is redeemable against the treasury with one transaction. Every payout leaves a receipt like this.

[Fab @Fab_RWA]

Mining needs $FAB. The wallet you mine with holds at least 50000 of $FAB and signs a 24-hour mining pass, a message, not a transaction.

The server checks the balance on chain. That is the reason to hold $FAB: it is the key to the machine.

Everyone gets one free trial run first.

https://t.co/sjzCAm5qrb

[Fab @Fab_RWA]

The rate has a governor. At every six-hour settlement the fab values the treasury and checks the last 24 hours of payouts. Over 5% of holdings, the rate halves. Under 1%, it recovers. The current multiplier is printed on the treasury page.

[Fab @Fab_RWA]

How you get paid on FAB, step by step.

1. Run your GPU in the browser. 15 minutes minimum.
2. The fab checks the work and reads which companies built your chip.
3. You earn $0.10 per run plus $0.0009 for every 40 teraflops verified.
4. Every 6 hours that gets turned into shares at that hour's price: NVDA, TSM, SK hynix, whoever built your card.
5. You claim once. The stock tokens land in your wallet.

The only cost on the way out is the gas for that one claim. No cut, no fee. The treasury pays from what it holds.

https://t.co/pJQHiuO6g5

[Fab @Fab_RWA]

Sustained beats peak. On an Apple M2 Pro the memory test hits 183 GB/s in the first seconds and averages 27 GB/s over a 15-minute run, because the chip throttles. FAB scores the whole run, not the first second. The leaderboard is per GPU model.

[Fab @Fab_RWA]

A laptop pretending to be a 4090 gets paid like a laptop.

Each run starts with a signed token from the fab, has to come back after the full run length, 15, 30 or 60 minutes, once, and the score is capped at the ceiling for that vendor's silicon.

Four runs an hour per IP.

[Fab @Fab_RWA]

The treasury is a contract, not a promise

https://t.co/vOkFhp4m7M on Robinhood Chain holds the stock tokens FAB pays out and can only pay what it holds. Every deposit is a public event with a memo. Every claim is a transaction.

[Fab @Fab_RWA]

How a dollar moves through FAB.

A verified run earns $0.10 plus $0.009 per 40 teraflops.

Every six hours the fab converts what you are owed into shares at that hour's price and signs a voucher. You redeem it against the treasury and the stock tokens land in your wallet. 1% of fees go into the treasury as NVDA, TSM and SK hynix tokens.

[Fab @Fab_RWA]

Sustained beats peak. On an Apple M2 Pro the memory test hits 183 GB/s in the first seconds and averages 27 GB/s over a 15-minute run, because the chip throttles. FAB scores the whole run, not the first second. The leaderboard is per GPU model.

[Fab @Fab_RWA]

Got 60 seconds? That is the whole product.

Pick a test program, press begin, and the wafer fills while the card works. At 60 seconds the fab verifies what it saw, splits the reward by who built your silicon, and queues it for the next epoch.

Nothing installed.

[Fab @Fab_RWA]

We've paid the dex listing.

[Fab @Fab_RWA]

Every GPU is a supply chain.

This is what FAB reads when your run finishes: who designed the chip, which foundry made the silicon, who built the memory. That lineage is your reward map. On an @NVIDIA card it pays 48% NVDA, 31% TSM, 14% SK hynix, and the map changes with the hardware you bring.