Steam GPU Launch Console

Generate the commands to deploy a per-player Windows GPU cloud-gaming host on AWS — NVIDIA gaming driver, Amazon DCV browser streaming, Steam Offline Mode library on its own EBS volume, and IAM-scoped self-troubleshooting.

NO CREDENTIALS STORED · NO AWS CALLS MADE

Parameters

Keep it near you — a distant region also looks odd to Steam
Decrypts the local Administrator password
gp3, encrypted, survives stop/start
Must end in /32. Blank lets launch.sh detect it
One Windows instance = one player. Amazon DCV's Windows server supports console sessions only — virtual sessions and GPU sharing are Linux-only. Scale by adding instances, not by adding users to one box.
Steam credential sharing is outside Valve's terms. Offline Mode is legitimate for your own second machine. For other people, use Steam Families — up to six members, one play slot per owned copy, shared games playable offline.
Browser streaming is the convenience tier. The DCV web client is TCP-only and capped at 2 × 1920×1080. The native client gets QUIC over UDP 8443 and up to 4K.

Launch from your machine

Deploy with launch.sh

Validates first, then applies. No inbound RDP at any point.

Launch on Hugging Face compute

Validate only

Needs no AWS credentials at all. Run this first.

Deploy

Credentials travel as Hub secrets; nothing is written to the repo or image.

Operate

RDP over Systems Manager

Reaches 3389 without ever opening 3389.

Health report

The instance reports its own failures as a JSON verdict.

After the deploy

  1. Upload the Amazon DCV server MSI to C:\DCV on the instance and re-run the DCV section of bootstrap.ps1, then reboot so the GPU driver loads.
  2. Confirm the licence with nvidia-smi -q — expect NVIDIA Cloud Gaming and License Status : Licensed.
  3. Open https://<elastic-ip>:8443/ in a browser and connect to the console session.
  4. Configure Steam Offline Mode: save credentials, clear the download queue, launch each game once online, then Steam > Go Offline.
  5. Run the health report, then stop the instance between sessions — the GPU hour is the whole cost.