Phi-4 vs Gemma-2-9b-it
Signed, community-submitted decode tok/s, prefill, and TTFT for Phi-4 vs Gemma-2-9b-it — every number links to the run it came from.
Verdict
On the RTX 5090 (32GB), Gemma-2-9b-it decodes at 153 tok/s versus 141 tok/s for Phi-4, 1.1× faster. Across 3 hardware configs measured on both, Gemma-2-9b-it is faster on 3 of 3. Every cell in the table below links to the submitted run it came from.
Hardware with data on both models
| Hardware | Phi-4 decode | Gemma-2-9b-it decode | Δ | Source runs |
|---|---|---|---|---|
| RTX 5090 (32GB) | 140.9tok/s | 152.7tok/s | -11.8 | r_e-k4aea8ipr · r__b_bzmmab_8 |
| M3 Ultra (60-core GPU) | 74.37tok/s | 89.45tok/s | -15.1 | r_sqzp-0rdez- · r_iz137eqvuzy |
| M3 Pro (18-core GPU) | 15.61tok/s | 21.97tok/s | -6.4 | r_-w8hnn61va_ · r_q7t7b7dcuz5 |
See also: Phi-4 benchmarks · Gemma-2-9b-it benchmarks · All hardware · All models · Methodology