TokenMark own benchmark run: NVIDIA H100 SXM5 80GB, 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 H100 80GB HBM3, 580.105.08, 700.00 W, 1980 MHz, 5, 16, 81559 MiB
llama.cpp b11609, built from source with CUDA 12.8 for sm_90 (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: 171.69 tok/s, TTFT 87.1 ms, 270.1 W, 2 run(s)
  Qwen3.5-27B (unsloth/Qwen3.5-27B-GGUF) Q4_K_M: 68.51 tok/s, TTFT 268.1 ms, 451.7 W, 2 run(s)
  Qwen3.5-35B-A3B (unsloth/Qwen3.5-35B-A3B-GGUF) Q4_K_M: 185.18 tok/s, TTFT 125.7 ms, 248.9 W, 2 run(s)
  Qwen3.6-27B (Jackrong/Qwen3.6-27B-GGUF) Q4_K_M: 68.1 tok/s, TTFT 267.4 ms, 443.1 W, 2 run(s)
  Qwen3.6-27B (Jackrong/Qwen3.6-27B-GGUF) Q8_0: 60.23 tok/s, TTFT 267.8 ms, 465.5 W, 2 run(s)
  Qwen3.6-35B-A3B (unsloth/Qwen3.6-35B-A3B-GGUF) UD-Q4_K_M: 188.13 tok/s, TTFT 123.8 ms, 247.4 W, 2 run(s)
  Qwen3.6-35B-A3B (unsloth/Qwen3.6-35B-A3B-GGUF) UD-Q4_K_XL: 183.08 tok/s, TTFT 131.9 ms, 245 W, 2 run(s)
  Qwen3.6-35B-A3B (unsloth/Qwen3.6-35B-A3B-GGUF) UD-Q5_K_M: 185.29 tok/s, TTFT 124.8 ms, 250.6 W, 2 run(s)
  Qwen3.6-35B-A3B (unsloth/Qwen3.6-35B-A3B-GGUF) UD-Q6_K: 178.15 tok/s, TTFT 127.1 ms, 246.3 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).