TokenMark own benchmark run: NVIDIA A100-SXM4-40GB, 2026-10-14

Rented from Lambda Cloud, 1 GPU. Measured by TokenMark, not a vendor or community figure.
nvidia-smi (name, driver, power limit, max SM clock, PCIe gen, PCIe width, memory): NVIDIA A100-SXM4-40GB, 570.148.08, 400.00 W, 1410 MHz, 4, 16, 40960 MiB
llama.cpp b11609, built from source with CUDA 12.8 for sm_80 (the rows' "CUDA" setting is the driver's CUDA version).
Benchmark: poor-pauls-benchmark (ppb) 2668a04ce007d8b7aecb2af2e896435cc2506cef, llama-server runner, no extra flags.
Settings: ctx 8192, batch 512 and 1024, 1 concurrent user, 10 ShareGPT prompts x 256 generated tokens.
tok/s = generated tokens / wall time per request, includes time to first token. Each row is the median
of its baseline runs (rows whose runs disagree by more than 25% are dropped), same rule as the ppb rows
from other users on tokenmark.app.

Rows:
  gemma-4-26B-A4B-it (unsloth/gemma-4-26B-A4B-it-GGUF) UD-Q8_K_XL: 99.91 tok/s, TTFT 121.8 ms, 174.7 W, 2 run(s)
  Qwen3.5-27B (unsloth/Qwen3.5-27B-GGUF) Q4_K_M: 42.38 tok/s, TTFT 393.2 ms, 296.3 W, 2 run(s)
  Qwen3.5-35B-A3B (unsloth/Qwen3.5-35B-A3B-GGUF) Q4_K_M: 122.76 tok/s, TTFT 174.1 ms, 171.6 W, 2 run(s)
  Qwen3.6-27B (Jackrong/Qwen3.6-27B-GGUF) Q4_K_M: 42.39 tok/s, TTFT 391 ms, 294 W, 2 run(s)
  Qwen3.6-27B (Jackrong/Qwen3.6-27B-GGUF) Q8_0: 34.09 tok/s, TTFT 396.2 ms, 286.7 W, 2 run(s)
  Qwen3.6-35B-A3B (unsloth/Qwen3.6-35B-A3B-GGUF) UD-Q4_K_M: 123.99 tok/s, TTFT 174.9 ms, 169.6 W, 2 run(s)
  Qwen3.6-35B-A3B (unsloth/Qwen3.6-35B-A3B-GGUF) UD-Q4_K_XL: 122.11 tok/s, TTFT 173.5 ms, 171.1 W, 2 run(s)
  Qwen3.6-35B-A3B (unsloth/Qwen3.6-35B-A3B-GGUF) UD-Q5_K_M: 120.89 tok/s, TTFT 175.8 ms, 173.1 W, 2 run(s)
  Qwen3.6-35B-A3B (unsloth/Qwen3.6-35B-A3B-GGUF) UD-Q6_K: 118.58 tok/s, TTFT 177.1 ms, 179 W, 2 run(s)

results.jsonl: every ppb result row from this run, one per line, in the flat column layout of the
public ppb dataset (made with ppb's own flattener at the SHA above).