Perplexity Portable Computer on DGX Spark Simulator

Unified VRAM Partitioning (91.5 / 128.0 GB) 100% Zero-Cloud Sovereign

Orchestrator (42.0 GB)

Subagents (25.5 GB)

KV Cache (24.0 GB)

Sandbox/DB (0.0 GB)

Free Headroom (36.5 GB)

Total Allocated

91.5 GB

Free Headroom

36.5 GB

Orch Token Rate

48.2 tok/s

Subagent Agg Rate

134.6 tok/s

Cloud Dependency

Hardware Allocation Parameters

Orchestrator Model & Quantization 70B Q4

Parallel Subagent Workers 3 Agents

Subagent Model & Quantization 8B Q8 (8.5 GB/ea)

Active Context Window Horizon 32k Tokens (24.0 GB KV)

Local Sandbox & Vector Store 0 GB (Vector/Code Sandbox)

DGX Spark PCIe Unified bus: 2,048 GB/s bandwidth. 0 egress packets detected.

Local Agent Execution DAG

1. Orchestrator Task Decomposition 70B Q4 @ 48.2 tok/s

3. Local Vector Search & Code Sandbox Isolated Memory Ring

4. Unified Orchestrator Synthesis Zero-Cloud Local Stream

System initialized. Click "Run Agent Pipeline" or adjust hardware parameters to benchmark local DGX Spark capacity.