{
  "version": 1,
  "einheiten": "Decimal: 1 GB = 1,000,000,000 bytes; 1 TB = 1,000 GB.",
  "hardware_heute": "NVIDIA GeForce RTX 4080 (16 GB VRAM)",
  "hardware_plan": "Two Mac Studio machines, 512 GB each; not delivered.",
  "rest_hinweis": "Budget minus model files leaves an upper bound for context, runtime buffers and the operating system; no context length or runtime compatibility is promised.",
  "slug": "qwen3-5-9b-q4-k-m",
  "anzeige": "Qwen3.5 9B",
  "hf_repo": null,
  "variante": "Q4_K_M",
  "groesse": {
    "bytes": null,
    "gb_gerundet": "6.55",
    "dateien": null,
    "quelle": "MESSUNG_2026-10-13.json",
    "abgerufen": "2026-10-26",
    "hinweis": "Local model-file size recorded on 13 October 2026, rounded to 0.01 GB; exact bytes and Hugging Face repository were not recorded."
  },
  "budgets": [
    {
      "id": "24gb",
      "anzeige": "24 GB",
      "plan": false,
      "hinweis": "Comparison budget; the measured RTX 4080 has 16 GB VRAM.",
      "passt": true,
      "rest_bytes": "17450000000",
      "gerundete_grundlage": true
    },
    {
      "id": "512gb",
      "anzeige": "512 GB",
      "plan": true,
      "hinweis": "One planned Mac Studio; not delivered.",
      "passt": true,
      "rest_bytes": "505450000000",
      "gerundete_grundlage": true
    },
    {
      "id": "1tb",
      "anzeige": "1 TB",
      "plan": true,
      "hinweis": "Conservative comparison budget across two planned 512 GB Macs; not delivered. Not one shared memory pool.",
      "passt": true,
      "rest_bytes": "993450000000",
      "gerundete_grundlage": true
    }
  ],
  "messungen": [
    {
      "id": "MESSUNG_2026-10-13",
      "quelle": "MESSUNG_2026-10-13.json",
      "zeit": "2026-10-13T13:30:09",
      "zeitzone": null,
      "hardware": "NVIDIA GeForce RTX 4080",
      "vram_gesamt_mib": 16376,
      "software": "LM Studio",
      "software_version": null,
      "kontext_tokens": null,
      "methode": {
        "temperatur": null,
        "max_tokens": 300,
        "wiederholungen": 2,
        "aggregation": "recorded medians",
        "prompts": "Fresh identifier and topic per request; no prompt cache.",
        "exakte_prompts": null
      },
      "laeufe": [
        {
          "id": "kurz",
          "tok_s": 92.7,
          "ttft_s": 0.08,
          "prompt_tokens": 32,
          "watt": 155.5,
          "tok_s_einzeln": [
            92.1,
            93.4
          ]
        },
        {
          "id": "mittel",
          "tok_s": 94.5,
          "ttft_s": 0.3,
          "prompt_tokens": 855.5,
          "watt": 163.6,
          "tok_s_einzeln": [
            94.6,
            94.3
          ]
        },
        {
          "id": "lang",
          "tok_s": 93.3,
          "ttft_s": 0.74,
          "prompt_tokens": 3432.5,
          "watt": 174.5,
          "tok_s_einzeln": [
            93.2,
            93.5
          ]
        }
      ],
      "eur_pro_mio_tokens": 0.16,
      "strompreis_eur_kwh": 0.35,
      "stromumfang": "GPU only",
      "vram_belegt_gb": 6.9,
      "rohurl": "/lab/qwen3-5-9b-q4-k-m/data.json"
    },
    {
      "id": "MESSUNG_2026-10-27",
      "quelle": "MESSUNG_2026-10-27.json",
      "zeit": "2026-10-27T13:30:02+01:00",
      "zeitzone": "recorded in timestamp",
      "hardware": "NVIDIA GeForce RTX 4080",
      "vram_gesamt_mib": 16376,
      "software": "LM Studio",
      "software_version": "2.61.0",
      "kontext_tokens": 16384,
      "methode": {
        "temperatur": null,
        "max_tokens": 1500,
        "wiederholungen": 2,
        "aggregation": "recorded medians",
        "prompts": "Fresh identifier and topic per request; no prompt cache.",
        "exakte_prompts": null
      },
      "laeufe": [
        {
          "id": "kurz",
          "tok_s": 93.1,
          "ttft_s": 0.09,
          "prompt_tokens": 35,
          "watt": 234.8,
          "tok_s_einzeln": [
            92.9,
            93.2
          ],
          "ttft_wand_s": 2.13,
          "ttft_wand_methode": "hergeleitet: dauer_s - gen_s (stream:false)",
          "ttft_engine_s": 0.09,
          "kosten_eur_1m_tokens": "0.2451963241436925647451963241",
          "strom_leistungsfeld": "power.draw.instant",
          "tokenbasis": "alle erzeugten Tokens einschliesslich Reasoning (completion_tokens)",
          "antwort_tokens": 1500,
          "reasoning_tokens": 1195.5,
          "sichtbare_tokens": 304.5
        },
        {
          "id": "mittel",
          "tok_s": 93.2,
          "ttft_s": 0.33,
          "prompt_tokens": 868.5,
          "watt": 240,
          "tok_s_einzeln": [
            93.2,
            93.3
          ],
          "ttft_wand_s": 2.13,
          "ttft_wand_methode": "hergeleitet: dauer_s - gen_s (stream:false)",
          "ttft_engine_s": 0.33,
          "kosten_eur_1m_tokens": "0.2503576537911301859799713877",
          "strom_leistungsfeld": "power.draw.instant",
          "tokenbasis": "alle erzeugten Tokens einschliesslich Reasoning (completion_tokens)",
          "antwort_tokens": 1500,
          "reasoning_tokens": 1494,
          "sichtbare_tokens": 6
        },
        {
          "id": "lang",
          "tok_s": 91.9,
          "ttft_s": 0.7,
          "prompt_tokens": 3442,
          "watt": 242.1,
          "tok_s_einzeln": [
            91.9,
            91.9
          ],
          "ttft_wand_s": 2.19,
          "ttft_wand_methode": "hergeleitet: dauer_s - gen_s (stream:false)",
          "ttft_engine_s": 0.7,
          "kosten_eur_1m_tokens": "0.2561207834602829162132752992",
          "strom_leistungsfeld": "power.draw.instant",
          "tokenbasis": "alle erzeugten Tokens einschliesslich Reasoning (completion_tokens)",
          "antwort_tokens": 1500,
          "reasoning_tokens": 1500,
          "sichtbare_tokens": 0
        }
      ],
      "eur_pro_mio_tokens": 0.25,
      "strompreis_eur_kwh": 0.35,
      "stromumfang": "GPU only",
      "vram_belegt_gb": 6.9,
      "rohurl": "/lab/qwen3-5-9b-q4-k-m/data.json",
      "software_version_quelle": "engine",
      "engine": "llama.cpp-win-x86_64-nvidia-cuda12-avx2",
      "engine_version": "2.61.0",
      "strom_leistungsfeld": "power.draw.instant",
      "tokenmethode": "native stats.tokens_per_second; alle erzeugten Tokens einschliesslich Reasoning (completion_tokens)"
    }
  ]
}