FLEET HARDWARE OBSERVATORY V13. 7 silos, 140+ cores, 336GB RAM, 168GB VRAM.
Each silo is a quasicrystal tile. Load heatmap updates at 337ms. pcdev RTX 5090 32GB = fleet's deepest reasoning pool.
Route heavy inference there. forge disk C: 95% is a live critical alert.
§1 — FLEET RESOURCE SUMMARY V13
95%
forge C: Disk (CRITICAL)
32 GB
pcdev VRAM (fleet max)
⚠ DISK CRITICAL — forge 192.168.2.12 — C: drive at 95%
Action required: docker system prune -af on forge. solvency degraded to ~0.15.
Do not dispatch heavy Docker build jobs to forge until resolved.
§2 — SILO DETAIL + QS READINESS
msi01
192.168.2.18 · Belt64 Seg 8 · QS weight 14.134
CPU: Core Ultra 9 275HX 24c
GPU: RTX 5090 Laptop 24GB
RAM: 64GB DDR5 (WSL 32GB)
Role: AKS control, campfire primary, CoreDNS
QS: QS-READY
forge
192.168.2.12 · Belt64 Seg 11 · QS weight 7.067
CPU: i9-14900KS 24c
GPU: RTX 4090 24GB
RAM: 64GB DDR5
Role: GPU inference, Docker builds (DISK CRITICAL)
QS: DISK-DEGRADED
msclo
192.168.2.19 · Belt64 Seg 9 · QS weight 14.134
CPU: Core Ultra 9 275HX 24c
GPU: RTX 5090 24GB
RAM: 64GB DDR5
Role: CLO/legal workloads, yLAW wiki, ARB
QS: QS-READY
yone
192.168.2.23 · Belt64 Seg 10 · QS weight 7.067
CPU: i9-14900KS 24c
GPU: RTX 5080 16GB
RAM: 64GB DDR5
Role: PEMCLAU, LHVCP k3d, Ollama inference
QS: QS-READY (LOCO dispatch host)
pcdev
192.168.2.16 · Belt64 Seg 13 · QS weight 4.711
CPU: i9-9900K 8c
GPU: RTX 5090 32GB ← FLEET MAX VRAM
RAM: 16GB (constraint)
Role: Deep reasoning, Lean4, joffe-math, RL training
QS: QS-READY GPU inference priority
lounge
192.168.50.175 · Belt64 Seg 14 · QS weight 4.711
CPU: i9-12900K 16c
GPU: RTX 4090 24GB
RAM: 32GB DDR5
Role: Rendering, visual preview, secondary inference
QS: QS-READY
lilo
Tailscale 100.97.143.89 · Belt64 Seg 12 · QS weight 7.067
CPU: Core Ultra 9 275HX 24c
GPU: RTX 5090 Laptop 24GB
RAM: 32GB DDR5
Role: Family/creative, Namir E.coli, lilo-fleet
QS: TAILSCALE-QS (~10ms latency)
§3 — GPU OCEAN — 168GB TOTAL VRAM
pcdev RTX 5090 32GB = fleet's deepest reasoning pool. Route heavy inference, RL training, Lean4 proof search here.
Never route to msi01 Arc-only GPU paths. lounge RTX 4090 = rendering priority.
GPU routing rule: target_silo = argmax(vram_free_gb × composite_weight)
pcdev free (current): ~22GB VRAM · composite ≈ 4.711 × 0.7 = 3.30
forge free (current): ~11GB VRAM (disk degraded) · composite ≈ 7.067 × 0.3 = 2.12
msclo free (current): ~16GB VRAM · composite ≈ 14.134 × 0.65 = 9.19 ← next GPU dispatch
§4 — NETWORK TOPOLOGY
LAN-A — 192.168.2.x
msi01: 192.168.2.18
forge: 192.168.2.12
msclo: 192.168.2.19
yone: 192.168.2.23
pcdev: 192.168.2.16
NAS: 192.168.2.20
Speed: 1Gbps+
LAN-B — 192.168.50.x
lounge: 192.168.50.175
steamdeck: 192.168.50.193
Separate switch
Speed: 1Gbps
Latency: ~0.1ms
TAILSCALE + AKS
lilo: 100.97.143.89
msclo TS: 100.70.197.54
AKS: Azure API
TS latency: ~10ms
AKS: Tailscale-optional
§5 — V13 QS COMPOSITE WEIGHT FORMULA
silo_weight_composite = (1 - wpa_pressure) × vram_free_pct × cpu_free_pct × adelic_weight
DRG gate threshold: composite > 0.3 required
forge disk alert: disk_pressure=0.95 → composite multiplied by (1 - 0.95) = 0.05 penalty
γ₁ = 14.134725141734693 · day97-v131 · EOSE Labs · HWMON V13 Bonixer