NAVER × NVIDIA · Infrastructure
NVIDIA DSX Platform Architecture
Software-defined power and compute orchestration for gigawatt-scale AI factories
Core Layer
DSX OS — Distributed Lifecycle Management
Open, bare-metal operating system for multi-tenant AI factories. Provides automated hardware lifecycle control, zero-downtime self-healing (Remediation), and fleet-wide intelligence across all GPU racks.
Power Control
DSX MaxLPS
Dynamic power limit policy engine. Monitors GPU workload patterns in real time — reclaims stranded power from idle units and redistributes it to active inference loads.
+40% GPU Density
MW↑ Token Throughput
Grid Response
DSX Flex
Grid-responsive compute orchestrator. Syncs data center power draw with utility load-shedding signals and renewable energy availability, adjusting training queues in nanoseconds.
Nanosecond Response
Renewable-Aware
IT Signals (Software)
GPU temperature spikes, compute queue depth, inference latency thresholds
OT Signals (Facility)
Cooling pump flow rates, CRAC valve states, power bus voltage readings
Closed-Loop Action
GPU thermal event → cooling pump RPM adjustment in real time via MQTT
Result
Physical infrastructure responds to compute load without human intervention
Source: NVIDIA DSX Platform Technical Playbook · Warren Barkley, Chris Alexiuk (NVIDIA Infrastructure Engineering)