Source: https://docs.sogni.ai/run-a-worker/fast-worker/about-sogni-fast-worker/ # 🐝About Sogni Fast Worker Sogni Fast Worker is a Docker-based service for contributing supported NVIDIA GPU capacity to the Sogni Supernet. It can run on a dedicated Windows or Linux machine, on an NVIDIA DGX Spark, on an otherwise-idle personal system, or on a GPU hosting platform that supports custom containers. After installation, use [dashboard.sogni.ai/fast-workers](https://dashboard.sogni.ai/fast-workers/) to check status, configure downtime alerts, inspect jobs and earnings, review health telemetry, and change settings supported by the worker. See [Using the Workers Dashboard](https://docs.sogni.ai/run-a-worker/fast-worker/worker-dashboard/) for a complete tour. ## [#](https://docs.sogni.ai/run-a-worker/fast-worker/about-sogni-fast-worker/#choose-a-worker-type)Choose a worker type There are three current Fast Worker types, plus one two-machine deployment: | Worker | Best fit | Configuration | | --- | --- | --- | | **Comfy Worker** | Current image, video, music, editing, and Comfy-hosted workflows | Dashboard settings begin with 1.0.189; use the latest supported release for the current catalog and live-apply behavior; `.env` is also supported | | **Stable Diffusion Worker** | Stable Diffusion community models and ControlNet workloads | `.env` | | **LLM Worker** | Supported language and vision-language inference | Deployment environment | | **DGX Spark DeepSeek Pair** | DeepSeek V4 Flash Vision Exp on two cabled NVIDIA DGX Sparks, as one worker | Pair kit settings file; see [Run DeepSeek V4 Flash on a DGX Spark pair](https://docs.sogni.ai/run-a-worker/fast-worker/dgx-spark-deepseek-pair/) | Comfy Worker ships as two hardware families. Ordinary NVIDIA desktop and server cards run the x86-64 image; an **NVIDIA DGX Spark** runs a separate Linux ARM64 build with its own version numbers, its own release history, and a deliberately smaller set of qualified workflows. The installer detects a DGX Spark and selects that image on its own. See the [Comfy Worker (DGX Spark) release notes](https://docs.sogni.ai/run-a-worker/fast-worker/release-notes/comfy-worker-spark/) for what it serves and what the hardware requires. The former Flux Worker is retired. Do not install or reactivate it; current Flux-family workflows are served by Comfy Worker. Available models and their resource requirements change with the production catalog. [Which jobs your hardware can take](https://docs.sogni.ai/run-a-worker/fast-worker/hardware-requirements/) explains the GPU VRAM and host RAM each job type needs and how to set up a machine for the most jobs. Consult the live [Comfy workflow catalog](https://api.sogni.ai/v1/worker/config/comfy) or [Stable Diffusion model catalog](https://socket.sogni.ai/api/v1/config/models/nvidia) before sizing a machine. A generic GPU name or VRAM total does not guarantee eligibility for every workflow. You can change worker type by running the current installer again. A worker advertises only the workloads compatible with its type, GPU, cached files, and operator settings. ## [#](https://docs.sogni.ai/run-a-worker/fast-worker/about-sogni-fast-worker/#requirements)Requirements Before installing, provide: 1. A [Sogni account](https://app.sogni.ai/). Use one primary account and create a separate Fast Worker NFT token ID for each simultaneously running GPU worker. 2. A supported NVIDIA GPU, or an NVIDIA DGX Spark. Some current workloads begin at 16 GB of VRAM; others require 20 GB, 24 GB, 32 GB, or more. The worker filters the live catalog for the detected hardware, and a DGX Spark is limited further to the workflows qualified on that machine. 3. Docker Desktop using its Linux/WSL2 backend on Windows, or Docker Engine with NVIDIA Container Toolkit on Linux. A DGX Spark uses the Linux path on ARM64. 4. Enough host RAM and fast persistent storage for the chosen workflows. For MiniMax H3 and LTX video on a 24/32 GB GPU, the worker needs 47 GiB of usable RAM (a 50 GB Docker/WSL allocation to be safe); see [hardware requirements](https://docs.sogni.ai/run-a-worker/fast-worker/hardware-requirements/). Comfy model storage can reach several hundred gigabytes and changes as workflows are added or retired. Reserve additional space for Docker images, temporary downloads, logs, and the operating system. 5. Stable Internet access, safe power delivery, and physical cooling sized for sustained GPU compute. 6. The Sogni account API key and the unique Fast Worker NFT token ID. The interactive installer guides the initial authentication. GPU workloads can run for long periods when network demand is high. Sogni's telemetry, render-time rests, and temperature guard are best-effort operational aids—not a safety certification or a substitute for proper power, airflow, cooling, maintenance, vendor limits, and independent monitoring. The hardware owner/operator is responsible for safe installation and operation. Read [Protecting your hardware](https://docs.sogni.ai/run-a-worker/fast-worker/sogni-fast-worker-advanced-configuration/#protecting-your-hardware) before starting a worker. ## [#](https://docs.sogni.ai/run-a-worker/fast-worker/about-sogni-fast-worker/#how-earnings-work)How earnings work A connected worker advertises eligible workloads, receives jobs from the Supernet, and is credited for successfully completed work under the applicable payment path. - Regular Spark- and SOGNI-paid jobs credit worker earnings as jobs complete. - Unlimited subscription-covered jobs are an optional second path accounted through the monthly [Worker Subscription Revenue Share Pool](https://docs.sogni.ai/pricing/unlimited-plan-details/#9-worker-revenue-share). Review participation and results in [Subscription Earnings](https://dashboard.sogni.ai/subscription-earnings). For a dated, hardware-by-hardware outlook, read **[What an Nvidia GPU earns on Sogni in Sept 2026](https://blog.sogni.ai/blogs/what-a-gpu-earns-on-sogni/)**. It combines pay-as-you-go work and the Unlimited subscription pool using eight complete days of activity across 133 ComfyUI workers. The September scenarios make uptime and pool-growth assumptions explicit; they are forecasts before operating costs, not settled payouts. Demand, job mix, worker speed, availability, model eligibility, network policy, payment values, and operating costs all vary. Historical dashboard or leaderboard results are not a promise of future workload, revenue, profit, or cost recovery. ## [#](https://docs.sogni.ai/run-a-worker/fast-worker/about-sogni-fast-worker/#start-here)Start here [⬇️Running Sogni Fast Worker Locally](https://docs.sogni.ai/run-a-worker/fast-worker/running-sogni-fast-worker-locally/)[☁️Running Sogni Fast Worker Remotely](https://docs.sogni.ai/run-a-worker/fast-worker/running-sogni-fast-worker-remotely/)[Using the Workers Dashboard](https://docs.sogni.ai/run-a-worker/fast-worker/worker-dashboard/)[🧠Sogni Fast Worker Advanced Configuration](https://docs.sogni.ai/run-a-worker/fast-worker/sogni-fast-worker-advanced-configuration/)[❓Fast Worker FAQ](https://docs.sogni.ai/run-a-worker/fast-worker/fast-worker-faq/) ## [#](https://docs.sogni.ai/run-a-worker/fast-worker/about-sogni-fast-worker/#terms-and-support)Terms and support Operating a worker is governed by the [Sogni Terms of Service](https://www.sogni.ai/terms-of-service) and [Privacy Policy](https://www.sogni.ai/privacy-policy). The Terms include the controlling provisions for Worker operation, operator responsibilities, eligibility, rewards, warranties, and liability; this setup guide does not replace them. For help, email [app@sogni.ai](mailto:app@sogni.ai) or join the [Sogni Discord](https://discord.com/invite/2JjzA2zrrc).