Expert Residency in Motion
MoE Router & Dynamic Tiered Memory Subsystem
Pause
VRAM Pressure: OFF
Reset
Tier 1: VRAM [Hot · ~0.2ms]
Tier 2: System RAM [Warm · ~2.5ms]
Tier 3: NVMe SSD [Cold · ~18ms]
CUDA GRAPH REPLAY LANE
TOKENS
PROMPT
ROUTER
CUDA GRAPH
COMBINE
TELEMETRY
VRAM HIT RATE:
100%
AVG LATENCY:
0.20 ms
RESIDENCY TIER:
2 V / 2 R / 2 S
GRAPH REPLAY:
ACTIVE