{
  "chart": "s2s-chart/1",
  "kind": "bars",
  "title": "An older desktop: ten cores, 48 GB of DDR4, six models at Q4_K_M and two at Q8_0 (2026-09-10)",
  "subtitle": "Tokens a second while writing on an older ten-core desktop, six models at Q4_K_M and two of them at Q8_0; a longer bar writes faster; an older engine build.",
  "unit": "tok/s",
  "series": [
    "Q4_K_M",
    "Q8_0"
  ],
  "rows": [
    {
      "label": "qwen3.5:9b",
      "notes": "",
      "values": [
        {
          "value": "6.96",
          "source": "/ram-speeds-compared/data/older-desktop/gpu-3080-m1-cpu-box.json derived.decode_tok_s.median 6.96 (Ollama 0.32.13, digest 6488c96fa5fa…)"
        },
        {
          "value": "4.39",
          "source": "/ram-speeds-compared/data/older-desktop/gpu-3080-x2-cpu-box.json derived.decode_tok_s.median 4.394 (Ollama 0.32.13, digest 441ec31e4d2a…)"
        }
      ]
    },
    {
      "label": "gemma4:12b",
      "notes": "",
      "values": [
        {
          "value": "4.85",
          "source": "/ram-speeds-compared/data/older-desktop/gpu-3080-m2-cpu-box.json derived.decode_tok_s.median 4.847 (Ollama 0.32.13, digest 4eb23ef187e2…)"
        },
        {
          "value": "3.03",
          "source": "/ram-speeds-compared/data/older-desktop/gpu-3080-x1-cpu-box.json derived.decode_tok_s.median 3.032 (Ollama 0.32.13, digest 41c402fdddc2…)"
        }
      ]
    },
    {
      "label": "phi4:14b",
      "notes": "Q8_0 not run",
      "values": [
        {
          "value": "4.29",
          "source": "/ram-speeds-compared/data/older-desktop/gpu-3080-m3-cpu-box.json derived.decode_tok_s.median 4.289 (Ollama 0.32.13, digest ac896e5b8b34…)"
        },
        {
          "value": null,
          "source": "Q8_0 was not run for this model on this machine (the kit holds no file)"
        }
      ]
    },
    {
      "label": "mistral-small3.2",
      "notes": "Q8_0 not run",
      "values": [
        {
          "value": "2.71",
          "source": "/ram-speeds-compared/data/older-desktop/gpu-3080-m4-cpu-box.json derived.decode_tok_s.median 2.71 (Ollama 0.32.13, digest 5a408ab55df5…)"
        },
        {
          "value": null,
          "source": "Q8_0 was not run for this model on this machine (the kit holds no file)"
        }
      ]
    },
    {
      "label": "gemma4:26b",
      "notes": "Older model build (digest 5571076f3d70…)",
      "values": [
        {
          "value": "12.80",
          "source": "/ram-speeds-compared/data/older-desktop/gpu-3080-m5-cpu-box.json derived.decode_tok_s.median 12.802 (Ollama 0.32.13, digest 5571076f3d70…)"
        },
        {
          "value": null,
          "source": "Q8_0 was not run for this model on this machine (the kit holds no file)"
        }
      ]
    },
    {
      "label": "Nemotron 30B-A3B",
      "notes": "Q8_0 not run",
      "values": [
        {
          "value": "13.06",
          "source": "/ram-speeds-compared/data/older-desktop/gpu-3080-m6-cpu-box.json derived.decode_tok_s.median 13.057 (Ollama 0.32.13, digest e7a64ff15fb1…)"
        },
        {
          "value": null,
          "source": "Q8_0 was not run for this model on this machine (the kit holds no file)"
        }
      ]
    }
  ],
  "source": "This page's kit, older-desktop/gpu-3080-m1…m6-cpu-box.json, -x1 and -x2, derived.decode_tok_s.median, three warm runs each under Ollama 0.32.13 on 2026-09-10; the thread count was not recorded. Table T10."
}