{
  "title": "Gemma 4 12B IT QAT Q4_0 on RTX 5090: long-context measurements",
  "test_date": "2026-09-06",
  "citation": "llm-speed (2026-09-06), Gemma 4 12B IT QAT Q4_0 on RTX 5090: long-context study. https://llm-speed.com/m/gemma-4-12b-it-qat#long-context",
  "snapshot_note": "Dated snapshot of nine accepted measurements; inspect linked runs for their current status. Estimated ctx targets are not actual prompt token counts.",
  "configuration": {
    "hardware": "RTX 5090 (32GB), AMD Ryzen 7 9850X3D, 30GB system RAM reported by WSL",
    "runtime": "llama.cpp CUDA",
    "runtime_commit": "9725a313be0528214c4a02fed906ddaf7b3f712e",
    "client_version": "0.0.5 with quantization/usage fixes",
    "client_source_commit": "140e9fea",
    "client_source_availability": "Internal revision; the exact client source is not available through the public CLI repository.",
    "suite": "suite-v1",
    "workload": "long-context-decay",
    "model_repository": "google/gemma-4-12B-it-qat-q4_0-gguf",
    "model_revision": "29d097773436b69ff9feafd636ab4cf873786537",
    "model_file": "gemma-4-12b-it-qat-q4_0.gguf",
    "model_sha256": "93567e57a8fe10b23569b9d9ec38cd005deedf71e29477c421a4b83f418a538b",
    "quantization": "Q4_0",
    "configured_context_tokens": 131072,
    "output_cap_tokens": 256,
    "parallel": 1,
    "gpu_layers": "49/49",
    "prompt_cache": false,
    "ram_cache_mib": 0,
    "thinking": "off",
    "temperature": 0,
    "seed": 42,
    "warm_model": true,
    "runtime_gpu_allocation_mib": 9685
  },
  "limits": [
    "Three sequential repetitions on one configuration; observed ranges are not confidence intervals.",
    "Not a full 128K-input or maximum-context test.",
    "No evaluation of answer quality or multi-user throughput.",
    "Other idle services remained GPU-resident. Allocation is not measured peak whole-device memory.",
    "Prompt targets use a four-characters-per-token estimate; prompt_tokens_actual is runtime-reported usage."
  ],
  "measurements": [
    {
      "repetition": 1,
      "source_url": "https://llm-speed.com/r/r_8q-uf-rq-gp",
      "ctx": 32000,
      "prompt_tokens_actual": 23872,
      "output_tokens": 98,
      "prefill_tps": 6022.356664978105,
      "decode_tps": 134.18110239592386,
      "ttft_ms": 3963.8967480659485,
      "wall_ms": 4687.196077108383,
      "p50_ms": 7.376587927341461,
      "p95_ms": 7.429484516382217,
      "skipped": false,
      "error": null
    },
    {
      "repetition": 1,
      "source_url": "https://llm-speed.com/r/r_8q-uf-rq-gp",
      "ctx": 64000,
      "prompt_tokens_actual": 47652,
      "output_tokens": 100,
      "prefill_tps": 6031.227868547562,
      "decode_tps": 131.71081555729006,
      "ttft_ms": 7900.878732919693,
      "wall_ms": 8652.868891954422,
      "p50_ms": 7.605286121368408,
      "p95_ms": 7.704463297128677,
      "skipped": false,
      "error": null
    },
    {
      "repetition": 1,
      "source_url": "https://llm-speed.com/r/r_8q-uf-rq-gp",
      "ctx": 128000,
      "prompt_tokens_actual": 95212,
      "output_tokens": 96,
      "prefill_tps": 4817.112236655199,
      "decode_tps": 115.04992286308963,
      "ttft_ms": 19765.368818998337,
      "wall_ms": 20592.928530931473,
      "p50_ms": 8.583517968654633,
      "p95_ms": 8.704683804512024,
      "skipped": false,
      "error": null
    },
    {
      "repetition": 2,
      "source_url": "https://llm-speed.com/r/r_1srqmpjnkgd",
      "ctx": 32000,
      "prompt_tokens_actual": 23872,
      "output_tokens": 98,
      "prefill_tps": 7161.038410828161,
      "decode_tps": 136.2665183827482,
      "ttft_ms": 3333.5947428941727,
      "wall_ms": 4045.8629829883575,
      "p50_ms": 7.291620969772339,
      "p95_ms": 7.560526788234711,
      "skipped": false,
      "error": null
    },
    {
      "repetition": 2,
      "source_url": "https://llm-speed.com/r/r_1srqmpjnkgd",
      "ctx": 64000,
      "prompt_tokens_actual": 47652,
      "output_tokens": 100,
      "prefill_tps": 6689.773186063769,
      "decode_tps": 137.27519506006652,
      "ttft_ms": 7123.111453056335,
      "wall_ms": 7845.051734924316,
      "p50_ms": 7.236937046051025,
      "p95_ms": 7.302313560247421,
      "skipped": false,
      "error": null
    },
    {
      "repetition": 2,
      "source_url": "https://llm-speed.com/r/r_1srqmpjnkgd",
      "ctx": 128000,
      "prompt_tokens_actual": 95212,
      "output_tokens": 96,
      "prefill_tps": 4784.301306689342,
      "decode_tps": 115.97619142027749,
      "ttft_ms": 19900.920509934425,
      "wall_ms": 20721.2485588789,
      "p50_ms": 8.521733522415161,
      "p95_ms": 8.64877837896347,
      "skipped": false,
      "error": null
    },
    {
      "repetition": 3,
      "source_url": "https://llm-speed.com/r/r_-lh-vczf7ik",
      "ctx": 32000,
      "prompt_tokens_actual": 23872,
      "output_tokens": 98,
      "prefill_tps": 7115.794497444604,
      "decode_tps": 142.61136023981624,
      "ttft_ms": 3354.7905309200287,
      "wall_ms": 4035.4317189455032,
      "p50_ms": 6.975228488445282,
      "p95_ms": 7.024567574262619,
      "skipped": false,
      "error": null
    },
    {
      "repetition": 3,
      "source_url": "https://llm-speed.com/r/r_-lh-vczf7ik",
      "ctx": 64000,
      "prompt_tokens_actual": 47652,
      "output_tokens": 100,
      "prefill_tps": 5394.69808247735,
      "decode_tps": 133.02551556115375,
      "ttft_ms": 8833.11712193489,
      "wall_ms": 9577.999054908752,
      "p50_ms": 7.473483502864838,
      "p95_ms": 7.53398295044899,
      "skipped": false,
      "error": null
    },
    {
      "repetition": 3,
      "source_url": "https://llm-speed.com/r/r_-lh-vczf7ik",
      "ctx": 128000,
      "prompt_tokens_actual": 95212,
      "output_tokens": 96,
      "prefill_tps": 4707.951429254146,
      "decode_tps": 123.65121664361361,
      "ttft_ms": 20223.658087968826,
      "wall_ms": 20993.955257058144,
      "p50_ms": 7.96955794095993,
      "p95_ms": 8.145021277666093,
      "skipped": false,
      "error": null
    }
  ],
  "summary": [
    {
      "target": 32000,
      "actual_prompt_tokens": 23872,
      "ttft_ms": {
        "median": 3354.7905309200287,
        "min": 3333.5947428941727,
        "max": 3963.8967480659485
      },
      "decode_tps": {
        "median": 136.2665183827482,
        "min": 134.18110239592386,
        "max": 142.61136023981624
      },
      "output_tokens": {
        "median": 98,
        "min": 98,
        "max": 98
      }
    },
    {
      "target": 64000,
      "actual_prompt_tokens": 47652,
      "ttft_ms": {
        "median": 7900.878732919693,
        "min": 7123.111453056335,
        "max": 8833.11712193489
      },
      "decode_tps": {
        "median": 133.02551556115375,
        "min": 131.71081555729006,
        "max": 137.27519506006652
      },
      "output_tokens": {
        "median": 100,
        "min": 100,
        "max": 100
      }
    },
    {
      "target": 128000,
      "actual_prompt_tokens": 95212,
      "ttft_ms": {
        "median": 19900.920509934425,
        "min": 19765.368818998337,
        "max": 20223.658087968826
      },
      "decode_tps": {
        "median": 115.97619142027749,
        "min": 115.04992286308963,
        "max": 123.65121664361361
      },
      "output_tokens": {
        "median": 96,
        "min": 96,
        "max": 96
      }
    }
  ]
}