Useful work for
compatible GPUs.
Different jobs need different models, software, memory and controls. IdleGPU is starting with one defined pilot and exploring other workloads as customer demand and testing justify them.
Text embeddings · current research pilot
Convert permitted public text into numerical representations for search and grouping. The installed worker uses a pinned MiniLM model. Returned output remains reported until independently checked; no commercial acceptance or earnings are enabled.
Understand the text pilot →Image processing · candidate workload
Examples include tagging, resizing and checking authorised public images. Each operation needs its own approved implementation and correctness checks. These jobs are not currently available through IdleGPU.
Transcription · candidate workload
Turn authorised audio into text using a model compatible with the worker. Quality, licensing and consent must be assessed. Private recordings and personal information are outside the current public-data pilot. This workload is not available yet.
Rendering · candidate workload
Produce images from approved scene assets on suitable installed workers. Software licensing, asset permissions, memory, isolation and output validation need separate testing. Rendering is not currently offered.
What makes a job eligible?
A defined input, an approved model or operation, compatible hardware, explicit provider opt-in and a measurable way to check the returned output. Customer-supplied executable code, confidential data and unrestricted remote access are outside the current pilot.
Batch processing means completing a collection of tasks with an agreed delivery window. It can apply to text, images or other workloads; it is not a restriction to embeddings. Real-time services and large-model training are not supported.
Browser and console contribution remain research only. A device name or browser graphics test does not prove model compatibility, physical GPU inference, sustained performance or permission for commercial use.