GPU vs CPU Comparison

Each dataset rendered side by side: CPU (left) and GPU (right). Same data, same weights, same alpha. GPU projections use float32 (vs CPU float64). Small precision differences change rank quantization sort order for tightly packed nodes, producing visually different but statistically equivalent layouts. Cluster structure and color grouping should match.

Initializing...

Karate Club — 34 nodes, group weight

Karate · group:8 · α=0 CPU

Karate · group:8 · α=0 GPU

Karate Club — topology α=0.75

Karate · balanced · α=0.75 CPU

Karate · balanced · α=0.75 GPU

Epstein — 364 nodes, edge types

Epstein · edgetype:8 · α=0 CPU

Epstein · edgetype:8 · α=0 GPU

Epstein — topology α=0.75

Epstein · group:5 edgetype:8 · α=0.75 CPU

Epstein · group:5 edgetype:8 · α=0.75 GPU

BitZoom Source — 433 nodes, kind weight

BZ Source · kind:8 group:3 · α=0 CPU

BZ Source · kind:8 group:3 · α=0 GPU

BitZoom Source — topology α=0.5

BZ Source · kind:8 group:3 · α=0.5 CPU

BZ Source · kind:8 group:3 · α=0.5 GPU

MITRE ATT&CK — 4736 nodes, property weighted

MITRE · platforms:6 killchain:4 · α=0 CPU

MITRE · platforms:6 killchain:4 · α=0 GPU

Email-EU — 1005 nodes, edge-only, topology

Email-EU · α=0.75 · rank quant CPU

Email-EU · α=0.75 · rank quant GPU