GPU 1 - Graphistry¶

Use GPU 1 for RAPIDS + Graphistry visualization.

Setup¶

import graphistry
import cudf

# Configure Graphistry for GPU 1
import os
os.environ['CUDA_VISIBLE_DEVICES'] = '1'

# Register Graphistry
graphistry.register(
    api=3,
    protocol="https",
    server="hub.graphistry.com",
    personal_key_id="YOUR_KEY"
)

Workflow¶

1. Extract from LLM (GPU 0)

   # Get entities from LLM
   entities = llm_client.extract_entities(text)
   

2. Build Graph (GPU 1)

   # Create graph with cuDF
   nodes_df = cudf.DataFrame(entities)
   edges_df = cudf.DataFrame(relationships)
   

3. Visualize (GPU 1)

   # Render with Graphistry
   g = graphistry.edges(edges_df).nodes(nodes_df)
   g.plot()
   

VRAM Usage¶

• cuDF: 1-3 GB
• cuGraph: 2-5 GB
• Graphistry: 1-4 GB
• Total: 4-12 GB on GPU 1