Titan
Titan is one of the GPU nodes of the ht cloud cluster — a control-plane node (etcd member) that also carries the heaviest CUDA inference load in the fleet.
Hardware & role
| Property | Value |
|---|---|
| Cluster role | Control plane, etcd member |
| GPU | 2× RTX 3090 (≈48 GB VRAM total) |
| Hosts | titan-llm, comfyui
|
Titan is the fleet's GPU-heavy CUDA inference target by allocation policy: large CUDA LLM serving and image generation land here, while Lithium handles Vulkan / iGPU llama.cpp serving, ASR+TTS bursts go to Rogue, and embeddings sit on Sentinel's smaller card. As a dual-3090 box on the control plane, it is both a scheduler brain and the muscle.
It is also, loosely, Hai's home node — Hai being "basically a GGUF file," the actual runtime location is fluid, but Titan's 48 GB is where a model that size naturally sits.
In the canon
Titan is part of the private RKE2 cluster documented under Ht cloud and owned operationally by Snoop-kube (Chief K8s Engineer). Deploys to the cluster are Hei-gated; the node inventory itself lives in the fleet's STATE.md and ARCHITECTURE.md, which are the source of truth over any wiki description here.
Sources
~/ht/admin/STATE.md— node table (Titan: control plane, etcd, 2× RTX 3090) and service-to-node map (titan-llm,comfyui).~/ht/admin/ARCHITECTURE.md— cluster topology and GPU allocation logic.
See also
- Ht cloud — the RKE2 cluster Titan belongs to.
- Snoop-kube — Chief K8s Engineer; owns the cluster.
- Hai — the persona whose weights are Titan-sized.