{
 "id": "llama-3.3-70b",
 "name": "Llama 3.3 70B",
 "repo": "meta-llama/Llama-3.3-70B-Instruct",
 "lab": {
  "name": "Meta",
  "country": "United States"
 },
 "released": "2024-11-26",
 "parameters": 70.55,
 "activeParameters": null,
 "maxContext": 131072,
 "architecture": "80 layers · GQA, 8 KV heads × 128",
 "page": "https://studiotvai.com/vram-requirements/llama-3-3-70b",
 "calculator": "https://studiotvai.com/llm-vram-calculator?model=llama-3.3-70b",
 "weightsGiB": {
  "bf16": 131.41,
  "fp8": 67.66,
  "q4_k_m": 39.88
 },
 "kvCacheGiBPerRequest": [
  {
   "context": 8192,
   "bf16": 2.5,
   "fp8": 1.25
  },
  {
   "context": 32768,
   "bf16": 10,
   "fp8": 5
  },
  {
   "context": 131072,
   "bf16": 40,
   "fp8": 20
  }
 ],
 "recurrentStateGiBPerRequest": 0,
 "gpusNeeded": {
  "scenario": "one request of 8192 tokens, BF16 KV cache; vLLM on data-center GPUs, llama.cpp elsewhere",
  "gpus": [
   {
    "gpu": "rtx-4090",
    "name": "GeForce RTX 4090",
    "memoryGB": 24,
    "engine": "llama.cpp",
    "byFormat": {
     "bf16": 7,
     "fp8": 4,
     "q4_k_m": 2
    }
   },
   {
    "gpu": "rtx-5090",
    "name": "GeForce RTX 5090",
    "memoryGB": 32,
    "engine": "llama.cpp",
    "byFormat": {
     "bf16": 5,
     "fp8": 3,
     "q4_k_m": 2
    }
   },
   {
    "gpu": "rtx-pro-6000",
    "name": "RTX PRO 6000 Blackwell",
    "memoryGB": 96,
    "engine": "llama.cpp",
    "byFormat": {
     "bf16": 2,
     "fp8": 1,
     "q4_k_m": 1
    }
   },
   {
    "gpu": "mac-m5-max-128",
    "name": "Mac M5 Max 128 GB",
    "memoryGB": 96,
    "engine": "llama.cpp",
    "byFormat": {
     "bf16": 2,
     "fp8": 1,
     "q4_k_m": 1
    }
   },
   {
    "gpu": "h100",
    "name": "NVIDIA H100",
    "memoryGB": 80,
    "engine": "vLLM",
    "byFormat": {
     "bf16": 2,
     "fp8": 2,
     "q4_k_m": 1
    }
   },
   {
    "gpu": "h200",
    "name": "NVIDIA H200",
    "memoryGB": 141,
    "engine": "vLLM",
    "byFormat": {
     "bf16": 2,
     "fp8": 1,
     "q4_k_m": 1
    }
   },
   {
    "gpu": "b200",
    "name": "NVIDIA B200 (HGX)",
    "memoryGB": 180,
    "engine": "vLLM",
    "byFormat": {
     "bf16": 1,
     "fp8": 1,
     "q4_k_m": 1
    }
   }
  ]
 },
 "ggufFiles": {
  "repo": "unsloth/Llama-3.3-70B-Instruct-GGUF",
  "gib": {
   "BF16": 131.43,
   "F16": 131.43,
   "IQ4_NL": 37.3,
   "IQ4_XS": 35.33,
   "Q2_K": 24.56,
   "Q2_K_L": 24.79,
   "Q3_K_M": 31.91,
   "Q3_K_S": 28.79,
   "Q4_0": 37.36,
   "Q4_1": 41.27,
   "Q4_K_M": 39.6,
   "Q4_K_S": 37.58,
   "Q5_K_M": 46.52,
   "Q5_K_S": 45.32,
   "Q6_K": 53.91,
   "Q8_0": 69.83,
   "UD-IQ1_M": 15.97,
   "UD-IQ1_S": 14.77,
   "UD-IQ2_M": 22.63,
   "UD-IQ2_XXS": 18.1,
   "UD-IQ3_XXS": 25.76,
   "UD-Q2_K_XL": 25.11,
   "UD-Q3_K_XL": 32.4,
   "UD-Q4_K_XL": 39.73,
   "UD-Q5_K_XL": 46.44,
   "UD-Q6_K_XL": 56.96,
   "UD-Q8_K_XL": 75.65
  }
 },
 "runsWith": [
  "vLLM",
  "SGLang",
  "llama.cpp (Ollama, LM Studio)",
  "MLX (Mac)"
 ],
 "verified": {
  "gguf": {
   "day": "2026-10-10",
   "repo": "unsloth/Llama-3.3-70B-Instruct-GGUF",
   "kv": "match",
   "ours": 327680,
   "theirs": 327680
  }
 },
 "updated": "2026-10-10T15:05:26.095Z"
}