{
  "license": "CC BY 4.0",
  "attribution": "btensai.com",
  "hardware": [
    {
      "id": "rtx-3060-12",
      "name": "NVIDIA RTX 3060 12 GB",
      "kind": "gpu",
      "memOptionsGb": [
        12
      ],
      "bandwidthGBs": 360
    },
    {
      "id": "rtx-5060-ti-16",
      "name": "NVIDIA RTX 5060 Ti 16 GB",
      "kind": "gpu",
      "memOptionsGb": [
        16
      ],
      "bandwidthGBs": 448
    },
    {
      "id": "rtx-4070-ti-super",
      "name": "NVIDIA RTX 4070 Ti Super 16 GB",
      "kind": "gpu",
      "memOptionsGb": [
        16
      ],
      "bandwidthGBs": 672
    },
    {
      "id": "rtx-3090",
      "name": "NVIDIA RTX 3090 24 GB (used)",
      "kind": "gpu",
      "memOptionsGb": [
        24
      ],
      "bandwidthGBs": 936
    },
    {
      "id": "rtx-3090-x2",
      "name": "2x NVIDIA RTX 3090 48 GB (used, layer split)",
      "kind": "gpu",
      "memOptionsGb": [
        48
      ],
      "bandwidthGBs": 936
    },
    {
      "id": "rtx-4090",
      "name": "NVIDIA RTX 4090 24 GB",
      "kind": "gpu",
      "memOptionsGb": [
        24
      ],
      "bandwidthGBs": 1008
    },
    {
      "id": "rtx-5090",
      "name": "NVIDIA RTX 5090 32 GB",
      "kind": "gpu",
      "memOptionsGb": [
        32
      ],
      "bandwidthGBs": 1792
    },
    {
      "id": "rtx-pro-6000",
      "name": "NVIDIA RTX Pro 6000 96 GB",
      "kind": "gpu",
      "memOptionsGb": [
        96
      ],
      "bandwidthGBs": 1792
    },
    {
      "id": "apple-m4",
      "name": "Apple M4 / Mac mini M4",
      "kind": "unified",
      "memOptionsGb": [
        16,
        24,
        32
      ],
      "bandwidthGBs": 120,
      "apple": true
    },
    {
      "id": "apple-m4-pro",
      "name": "Apple M4 Pro",
      "kind": "unified",
      "memOptionsGb": [
        24,
        48,
        64
      ],
      "bandwidthGBs": 273,
      "apple": true
    },
    {
      "id": "apple-m4-max",
      "name": "Apple M4 Max (40-core GPU)",
      "kind": "unified",
      "memOptionsGb": [
        48,
        64,
        128
      ],
      "bandwidthGBs": 546,
      "apple": true
    },
    {
      "id": "apple-m4-max-32",
      "name": "Apple M4 Max (32-core GPU)",
      "kind": "unified",
      "memOptionsGb": [
        36
      ],
      "bandwidthGBs": 410,
      "apple": true
    },
    {
      "id": "apple-m1-ultra",
      "name": "Apple M1 Ultra",
      "kind": "unified",
      "memOptionsGb": [
        64,
        128
      ],
      "bandwidthGBs": 800,
      "apple": true
    },
    {
      "id": "apple-m2-ultra",
      "name": "Apple M2 Ultra",
      "kind": "unified",
      "memOptionsGb": [
        64,
        128,
        192
      ],
      "bandwidthGBs": 800,
      "apple": true
    },
    {
      "id": "apple-m3-ultra",
      "name": "Apple M3 Ultra",
      "kind": "unified",
      "memOptionsGb": [
        96,
        256,
        512
      ],
      "bandwidthGBs": 819,
      "apple": true
    },
    {
      "id": "strix-halo-128",
      "name": "AMD Strix Halo (Ryzen AI Max+ 395)",
      "kind": "unified",
      "memOptionsGb": [
        32,
        64,
        96,
        128
      ],
      "bandwidthGBs": 256,
      "share": 0.75
    },
    {
      "id": "dgx-spark",
      "name": "NVIDIA DGX Spark",
      "kind": "unified",
      "memOptionsGb": [
        128
      ],
      "bandwidthGBs": 273,
      "share": 0.9
    },
    {
      "id": "cpu-ddr5",
      "name": "CPU only, dual-channel DDR5-5600",
      "kind": "cpu",
      "memOptionsGb": [
        16,
        32,
        64,
        96,
        128
      ],
      "bandwidthGBs": 89.6,
      "share": 0.8
    },
    {
      "id": "cpu-ddr4",
      "name": "CPU only, dual-channel DDR4-3200",
      "kind": "cpu",
      "memOptionsGb": [
        16,
        32,
        64,
        128
      ],
      "bandwidthGBs": 51.2,
      "share": 0.8
    }
  ],
  "models": [
    {
      "id": "llama-3.2-1b",
      "name": "Llama 3.2 1B",
      "totalParamsB": 1.24,
      "activeParamsB": 1.24,
      "layers": 16,
      "kvHeads": 8,
      "headDim": 64,
      "moe": false
    },
    {
      "id": "llama-3.2-3b",
      "name": "Llama 3.2 3B",
      "totalParamsB": 3.21,
      "activeParamsB": 3.21,
      "layers": 28,
      "kvHeads": 8,
      "headDim": 128,
      "moe": false
    },
    {
      "id": "mistral-7b",
      "name": "Mistral 7B v0.3",
      "totalParamsB": 7.25,
      "activeParamsB": 7.25,
      "layers": 32,
      "kvHeads": 8,
      "headDim": 128,
      "moe": false
    },
    {
      "id": "llama-3.1-8b",
      "name": "Llama 3.1 8B",
      "totalParamsB": 8.03,
      "activeParamsB": 8.03,
      "layers": 32,
      "kvHeads": 8,
      "headDim": 128,
      "moe": false
    },
    {
      "id": "qwen3-14b",
      "name": "Qwen3 14B",
      "totalParamsB": 14.8,
      "activeParamsB": 14.8,
      "layers": 40,
      "kvHeads": 8,
      "headDim": 128,
      "moe": false
    },
    {
      "id": "qwen3-32b",
      "name": "Qwen3 32B",
      "totalParamsB": 32.8,
      "activeParamsB": 32.8,
      "layers": 64,
      "kvHeads": 8,
      "headDim": 128,
      "moe": false
    },
    {
      "id": "qwen2.5-14b",
      "name": "Qwen2.5 14B",
      "totalParamsB": 14.8,
      "activeParamsB": 14.8,
      "layers": 48,
      "kvHeads": 8,
      "headDim": 128,
      "moe": false
    },
    {
      "id": "gemma-3-27b",
      "name": "Gemma 3 27B",
      "totalParamsB": 27,
      "activeParamsB": 27,
      "layers": 62,
      "kvHeads": 16,
      "headDim": 128,
      "localLayerFraction": 0.8333333333333334,
      "localWindow": 1024,
      "moe": false
    },
    {
      "id": "qwen3-30b-a3b",
      "name": "Qwen3 30B-A3B (MoE)",
      "totalParamsB": 30.5,
      "activeParamsB": 3.3,
      "layers": 48,
      "kvHeads": 4,
      "headDim": 128,
      "moe": true
    },
    {
      "id": "qwen2.5-32b",
      "name": "Qwen2.5 32B",
      "totalParamsB": 32.5,
      "activeParamsB": 32.5,
      "layers": 64,
      "kvHeads": 8,
      "headDim": 128,
      "moe": false
    },
    {
      "id": "mixtral-8x7b",
      "name": "Mixtral 8x7B (MoE)",
      "totalParamsB": 46.7,
      "activeParamsB": 12.9,
      "layers": 32,
      "kvHeads": 8,
      "headDim": 128,
      "moe": true
    },
    {
      "id": "llama-3.3-70b",
      "name": "Llama 3.3 70B",
      "totalParamsB": 70.6,
      "activeParamsB": 70.6,
      "layers": 80,
      "kvHeads": 8,
      "headDim": 128,
      "moe": false
    },
    {
      "id": "gpt-oss-120b",
      "name": "gpt-oss 120B (MoE)",
      "totalParamsB": 116.8,
      "activeParamsB": 5.1,
      "layers": 36,
      "kvHeads": 8,
      "headDim": 64,
      "localLayerFraction": 0.5,
      "localWindow": 128,
      "moe": true
    },
    {
      "id": "llama-3.1-405b",
      "name": "Llama 3.1 405B",
      "totalParamsB": 405,
      "activeParamsB": 405,
      "layers": 126,
      "kvHeads": 8,
      "headDim": 128,
      "moe": false
    }
  ],
  "quantizations": [
    {
      "id": "f16",
      "label": "FP16 / BF16",
      "bitsPerWeight": 16
    },
    {
      "id": "q8_0",
      "label": "Q8_0",
      "bitsPerWeight": 8.5
    },
    {
      "id": "q6_k",
      "label": "Q6_K",
      "bitsPerWeight": 6.56
    },
    {
      "id": "q5_k_m",
      "label": "Q5_K_M",
      "bitsPerWeight": 5.7
    },
    {
      "id": "q4_k_m",
      "label": "Q4_K_M",
      "bitsPerWeight": 4.85
    },
    {
      "id": "q3_k_m",
      "label": "Q3_K_M",
      "bitsPerWeight": 3.9
    },
    {
      "id": "q2_k",
      "label": "Q2_K",
      "bitsPerWeight": 3.35
    },
    {
      "id": "mxfp4",
      "label": "MXFP4",
      "bitsPerWeight": 4.25
    }
  ]
}