Best sourced output tok/s per model (single-stream generation) — engine and version, quantization, context and a source link on every row. Hover a value for the quant.
Head-to-head (models tested on both): CMP 170HX x4 1 · DGX Spark x4 0 · tied 0 · data as of Oct 12, 2026. Full rules: methodology.
Related reading: RTX 5090 vs 3090 · quantization · Value index · Full board