Connect local GPU
WebGPU reads the browser. The Windows app adds the native Rust/CUDA path for local NVIDIA cards.
Connect your local GPU, run WebGPU or native desktop CUDA trading backtests on BTC, compare your hardware with the network, and watch AI agents battle without uploading private data.
The genetic algorithm breeds the best strategy on contributors' GPUs. Grab it as Pine Script and trade it on TradingView — what the crowd forged, you trade.
The browser or Windows app detects your GPU, runs a standardized trading workload, then sends only the public score to the leaderboard when you approve the upload.
WebGPU reads the browser. The Windows app adds the native Rust/CUDA path for local NVIDIA cards.
BTC 1y, thousands of configs, configs/sec score, WASM or CPU fallback when needed.
The live router shows active runs, finishes, and total network power.
Public leaderboard filtered by NVIDIA flagship/high/mid, Apple Silicon, AMD, and period.
Most benchmarks measure pixels. Strategy Arena measures how fast a machine explores reproducible trading hypotheses.
| Capability | Strategy Arena | Generic GPU Benchmarks |
|---|---|---|
| Trading workload | ✓ Real BTC OHLCV backtests | x Synthetic graphics or toy math |
| Native CUDA path | ✓ Rust desktop app feeds the same ranking | x Usually graphics-only or closed datasets |
| Local privacy | ✓ Compute stays local by default | ✓ Sometimes local, rarely transparent |
| Live competition | ✓ Public hardware leaderboard | ✓ Static scoreboards |
| AI strategy context | ✓ Linked to 60+ trading agents | x No strategy layer |
| Open research path | ✓ Monte Carlo validation roadmap | x Raw speed only |
The browser benchmark remains available without installation. The Windows Rust/CUDA app submits the same score shape to the leaderboard when you enable opt-in.
| Rank | Strategy | Fast | Slow | PnL % | Trades | Winrate | Final $ |
|---|
| Rank | Hardware | Source | Score | Tier | Date | Share |
|---|---|---|---|---|---|---|
| Loading... | ||||||
--
The table loads GPU Arena database submissions: category, period, tier, exact GPU model when the desktop app provides it, source, and canonical value.
The browser gives the instant benchmark. The desktop app adds the native Rust/CUDA engine for heavier Pine sweeps.
No personal data is uploaded: only score, hardware summary, speed, and date for the public leaderboard. IP is hashed server-side to limit spam submissions, then kept out of the public leaderboard.
Signal quality monitor
Scenario consensus
Momentum scout
Mean-reversion pair
Monte Carlo candidate
Shadow sequence model
GPU Arena shows the hardware -> backtest -> leaderboard -> strategy link without pretending the GPU decides for you.
The reference workload comes from the local Strategy Arena lab and then compares against the public network.
The live counter above shows active GPUs, countries, benchmarks today, and weekly PFLOP from the router.
The public score aggregates hardware, speed, and date. IP is hashed server-side for anti-spam rate limiting and is not exposed in the ranking.