Show HN: ChartGPU – WebGPU-powered charting library (1M points at 60fps)

Original Article ↗ Hacker News Discussion ↗

Overall reception & perceived novelty

  • Many commenters find the demos visually impressive and very smooth, especially the million‑point example and candlestick charts.
  • Some consider this a strong candidate to replace existing “fast” charting libraries that choke around 100k–1M points.
  • Others argue the core idea (GPU‑accelerated charting) is not new, citing prior WebGL‑based libraries that handle millions to 100M+ points.

Performance, sampling, and data handling

  • Reported performance ranges from 30+ FPS on modest setups to refresh‑rate‑locked 165 FPS on high‑end GPUs.
  • Multiple people stress that downsampling (e.g. LTTB) can hide peaks and make statistics misleading; they request clear toggles and better documentation.
  • Several advocate columnar data layouts, typed arrays, and explicit Float32/Float64 support.
  • There’s a rich side discussion on strategies for huge datasets: adaptive sampling, min/max per bin, mip‑mapping, density/heatmap rendering, and even wavelet/DCT‑based approaches.

Browser support, security, and fallbacks

  • A recurring pain point is WebGPU availability: users on Linux, Firefox, Android, and Safari report needing flags, seeing blocklists, or having demos fail entirely.
  • Many request a WebGL or even 2D canvas fallback so charts remain usable without enabling experimental or privacy‑sensitive GPU features.
  • Some express strong concern that WebGPU is a “security/privacy nightmare” due to GPU driver reliability and fingerprinting surface; others see this as a tradeoff rather than a veto.

Bugs, UX issues, and rapid iteration

  • Several users report the data‑zoom slider and timeline scroll behaving unpredictably across macOS, Windows, and Firefox; panning thresholds and some buttons in the candlestick demo also misbehave.
  • The author responds quickly with fixes: corrected sliders, lower idle CPU usage via render‑on‑demand, improved candlestick streaming (up to millions of candles), and a benchmark mode toggle.

Architecture, integration, and use cases

  • Desired features include: OffscreenCanvas/worker‑thread rendering, zero‑copy data flow from workers, drawing/annotation tools, stacked area charts, graph/network visualization, Jupyter and React Native support, and potential integration as a backend for D3/Vega‑style grammars.
  • Suggested target markets are fintech (order book heatmaps, volatility surfaces, complex trading tools) and high‑density dashboards, though some argue many applications can rely on CPU plus good downsampling.

AI‑assisted development debate

  • Discovery of .cursor and .claude agent configs triggers a long meta‑thread: some dismiss the project as “AI slop,” others argue tools don’t matter if the output is solid.
  • There is discomfort with HN comments that appear LLM‑written, but also recognition that AI‑assisted coding is increasingly normal.