GPUWorks

About

GPUWorks is a GPU management and Model-as-a-Service platform where the server and web UI ship as a single static binary of about 40 MB, with state kept in one SQLite file. It pools bare-metal servers, rented cloud instances and Macs across NVIDIA, AMD, Ascend, PPU, Hygon and Apple Silicon, then estimates memory, picks GPUs and starts vLLM, SGLang or other engines behind one OpenAI-compatible endpoint. Its catalog lists 1,235 launch specs for 392 models.