A100 40GB PCIe llama.cpp b9935: gemma-4-31B-it UD-Q4_K_XL prefill
1197 tok/s prefill (ctx 0.5k, FA on); decode 41.2 t/s tg128
Fields
id
benchmarks/llamabench-a100pcie40-gemma4-31b-q4kxl-ppMetric
prefill_tok_s
Value
1,197
Unit
tok/s prefill (ctx 0.5k, FA on); decode 41.2 t/s tg128
Model
gemma-4-31B-it UD-Q4_K_XL
Methodology
llamabench.ai community row, reporter @wmeddie: llama.cpp build b9935, flash-attention on, KV cache f16, 0.5k context depth, no speculative decoding. Decode = tg128 single-stream (per-request, batch-1); prefill = site-reported figure. n=1 per model row — crowd-sourced, not a curated scoreboard.
Measured by
third_party