Share your GPU with Babel Mesh

How it runs

• Only while idle. By default sharing starts after your computer has been idle for 5 minutes and pauses when you come back. You can change that to share at any time, or change the wait.
• It steps aside when another program uses an NVIDIA card heavily. AMD and Intel cards do not report other programs' usage, so for them "the computer is idle" is the only signal.
• Outbound connections only, no inbound port. Your router needs no setting and you need no public address.
• It does not read your files. The client only downloads engines and models into its own folder, within a disk limit (20 GB by default, adjustable).
• You can limit video memory. Settings let you cap how much GPU memory is lent; the rest stays yours.
• No GPU? Lend the processor. A computer with no usable graphics card can share its CPU instead (small models and embeddings), using about half of its threads.

What you earn

Part of each request's price is credited to the account that provided the compute. Earned credits are first held for a few days while the checks continue; after that they can be spent on chat, images and video.

Credits above your own use may later be paid out in cash; that needs identity verification and review, and the date will be announced. You cannot withdraw cash yet.

What you need

System	Windows 10 or 11, Linux (x86-64), macOS 12 or later (Apple silicon, the GPU runs through Metal).
Graphics card	NVIDIA (CUDA), or AMD and Intel (Vulkan). Current models need about 5 to 12 GB of video memory; the more you have, the more work you can be given.
Network	An ordinary connection. The first run downloads about 0.4 GB of engine and a 1.1 GB speed-test model; after that, models as assigned, about 2 to 3 GB each.

Questions