Observed arrival · 2026-11-19
Synatra.space splits model layers across browser GPUs
The site presents a way to pool browser GPUs over WebRTC and run open-weight Hugging Face models across the connected devices.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- Teams and labs pooling browser GPUs for open models
- Worth noticing
- The page reports up to 73% less per-browser GPU memory when four peers share a 24-layer model.
Field notes
The described workflow is to host or join a password-protected room, load a model, and assign transformer layers to participating browser GPUs. The page says assignments rebalance as peers arrive or leave. Its console display lists room codes, per-device layer counts, memory figures, chat, and telemetry. The example rooms name Qwen2.5-0.5B-Instruct, Llama-3.2-1B-q4f16, and SmolLM2-360M-Instruct.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
◎NicheUnusually specific use
ƒJavaScriptBrowser-side code central
One card from the complete issue