// features · l1

L1 Blockchain

StackGPU runs a Ethereum-anchored attestation layer dedicated to verifiable GPU computation. Validators don't just settle payments — they re-execute and slash dishonest claims.

section · featureschain · Ethereum-anchored L1

Why a dedicated L1

Ethereum already settles billions of dollars in value every day, but it doesn't natively verify GPU computation. StackGPU's L1 is a specialized layer that commits state roots to Ethereum for finality, while running its own consensus optimized for compute-attestation throughput.

What the L1 does

  1. Receives inference attestations from hosts after each compute cycle. Each attestation contains: input hash, output hash, GPU model fingerprint, timestamp.
  2. Samples randomly a subset of cycles per epoch and assigns validators to re-execute.
  3. Compares outputs. If re-execution matches (within numerical tolerance), the host gets paid. If it diverges, slashing kicks in.
  4. Settles to Ethereum. Every epoch's state root is committed to a Ethereum smart contract for finality.

Architecture diagram

┌──────────────┐  inference job   ┌──────────────┐
│   Renter     │ ───────────────→ │   Host GPU   │
└──────────────┘   ←─ output ───  └──────┬───────┘
                                         │
                                  attestation
                                         │
                                         ▼
                              ┌──────────────────┐
                              │   StackGPU L1    │
                              │  (attestation,   │
                              │   sampling,      │
                              │   slashing)      │
                              └────────┬─────────┘
                                       │
                              epoch state roots
                                       ▼
                              ┌──────────────────┐
                              │   Ethereum mainnet │
                              │  (finality +     │
                              │   settlement)    │
                              └──────────────────┘

Throughput

Differences from EVM L1s

AspectEVM rollupStackGPU L1
Optimized forSmart contract stateCompute attestation
SettlementEthereum L1Ethereum mainnet
SamplingFraud proofs (week-long)Same-epoch re-execution
Block time2-12s~12s block cadence
No "rollup" in the marketing sense
We don't call this a rollup because we don't post tx data to Ethereum for replay — we post commitments and let validators dispute. The economic security model is closer to optimistic rollups with shorter dispute windows enabled by Ethereum finality.