The GPU pool.
Two GTX 1060 6GB cards are exclusive to a paid customer on the i7. The single i9 1060 is a shared Cloud + Lounge card: cloud acceleration when it helps, and Lounge AI/streaming when it is free.
Live facility load
Ledger 2026-11-01T06:24:00-06:00Passed through · host cannot sample GPU load
Host pollHost GPU · nouveau · idle
Host pollLive host load, memory, VM/container counts, and paid-VM status update every ~20 seconds from the Proxmox hosts. GPU utilization for the paid cards is not readable from the host because the 1060s are passed through to the VM; the telemetry column shows the host binding/runtime state instead.
12700KF · paid GPU node
Dedicated dual-GPU paid customer. Both cards are reserved for the paid VM; free workloads never preempt this node.
14900KF · Cloud + Lounge pool
One free GTX 1060 shared by Cloud acceleration (media transcoding, previews, AI helper) and Lounge workloads. Cloud acceleration is opportunistic and yields to AI/streaming. Plus gets first call when the pool is idle.
Workload priorities
JSON status →Cloud acceleration · where it helps
Shared i9 1060 · opportunisticCloud acceleration is deliberately not pinned to static web serving. It earns its keep on media transcode, GPU-rendered previews, AI helpers, and streamed playtests. It yields to AI/streaming, and paid dual-GPU capacity is never touched.
Allocation rules
- Paid dual-GPU customer has absolute priority on i7-118; two 1060s are exclusive to that VM.
- The free i9-230 1060 is the Lounge pool; nothing there preempts paid capacity.
- Only one exclusive workload may hold the i9 card at a time.
- Cloud acceleration is opportunistic and only if it can yield to AI/streaming.
- No GPU is sold as AI training capacity; these are fun-tier 6GB cards.
ASCS Credit is a stack-wide account balance for eligible Cloud orders and add-ons, including Lounge Plus. The same balance can cover paid GPU or Cloud services and Lounge Plus. The i9 1060 is not pinned to a Cloud service today because the current AMP/game-server stack has no GPU code path. It stays pooled for Lounge AI/streaming and opportunistic Cloud acceleration. Status is manually verified after hardware changes; the API is the coordination ledger, not a hypervisor.