{
 "id": "deepseek-v3.2",
 "name": "DeepSeek V3.2",
 "repo": "deepseek-ai/DeepSeek-V3.2",
 "lab": {
  "name": "DeepSeek",
  "country": "China"
 },
 "released": "2025-12-01",
 "parameters": 685.355329792,
 "activeParameters": 37,
 "maxContext": 163840,
 "architecture": "61 layers · MLA (512 + 64 latent) on 61 · indexer keys on 61",
 "page": "https://studiotvai.com/vram-requirements/deepseek-v3-2",
 "calculator": "https://studiotvai.com/llm-vram-calculator?model=deepseek-v3.2",
 "weightsGiB": {
  "native": 627.75,
  "bf16": 1251.47,
  "fp8": 627.46,
  "q4_k_m": 377.97
 },
 "kvCacheGiBPerRequest": [
  {
   "context": 8192,
   "bf16": 0.598,
   "fp8": 0.367
  },
  {
   "context": 32768,
   "bf16": 2.39,
   "fp8": 1.467
  },
  {
   "context": 131072,
   "bf16": 9.561,
   "fp8": 5.868
  },
  {
   "context": 163840,
   "bf16": 11.951,
   "fp8": 7.335
  }
 ],
 "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": {
     "native": 32,
     "bf16": 64,
     "fp8": 32,
     "q4_k_m": 24
    }
   },
   {
    "gpu": "rtx-5090",
    "name": "GeForce RTX 5090",
    "memoryGB": 32,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 24,
     "bf16": 48,
     "fp8": 24,
     "q4_k_m": 16
    }
   },
   {
    "gpu": "rtx-pro-6000",
    "name": "RTX PRO 6000 Blackwell",
    "memoryGB": 96,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 7,
     "bf16": 16,
     "fp8": 7,
     "q4_k_m": 5
    }
   },
   {
    "gpu": "mac-m5-max-128",
    "name": "Mac M5 Max 128 GB",
    "memoryGB": 96,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 7,
     "bf16": 16,
     "fp8": 7,
     "q4_k_m": 5
    }
   },
   {
    "gpu": "strix-halo",
    "name": "Ryzen AI Max+ 395 128 GB",
    "memoryGB": 96,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 7,
     "bf16": 16,
     "fp8": 7,
     "q4_k_m": 5
    }
   },
   {
    "gpu": "dgx-spark",
    "name": "DGX Spark (GB10)",
    "memoryGB": 128,
    "engine": "llama.cpp",
    "byFormat": {
     "native": 6,
     "bf16": 16,
     "fp8": 6,
     "q4_k_m": 3
    }
   },
   {
    "gpu": "h100",
    "name": "NVIDIA H100",
    "memoryGB": 80,
    "engine": "vLLM",
    "byFormat": {
     "native": 16,
     "bf16": 24,
     "fp8": 16,
     "q4_k_m": 6
    }
   },
   {
    "gpu": "h200",
    "name": "NVIDIA H200",
    "memoryGB": 141,
    "engine": "vLLM",
    "byFormat": {
     "native": 6,
     "bf16": 16,
     "fp8": 6,
     "q4_k_m": 4
    }
   },
   {
    "gpu": "b200",
    "name": "NVIDIA B200 (HGX)",
    "memoryGB": 180,
    "engine": "vLLM",
    "byFormat": {
     "native": 5,
     "bf16": 16,
     "fp8": 5,
     "q4_k_m": 3
    }
   }
  ]
 },
 "ggufFiles": {
  "repo": "unsloth/DeepSeek-V3.2-GGUF",
  "gib": {
   "BF16": 1250.09,
   "UD-TQ1_0": 150.2,
   "IQ4_NL": 353.1,
   "IQ4_XS": 333.74,
   "Q2_K": 228.51,
   "Q2_K_L": 228.72,
   "Q3_K_M": 298.24,
   "Q3_K_S": 270.5,
   "Q4_0": 353.99,
   "Q4_1": 391.87,
   "Q4_K_M": 377.56,
   "Q4_K_S": 354.89,
   "Q5_K_M": 443.46,
   "Q5_K_S": 430.86,
   "Q6_K": 513.45,
   "Q8_0": 664.3,
   "UD-IQ1_M": 185.62,
   "UD-IQ1_S": 171.48,
   "UD-IQ2_M": 212.44,
   "UD-IQ2_XXS": 202.22,
   "UD-IQ3_XXS": 254.37,
   "UD-Q2_K_XL": 229.67,
   "UD-Q3_K_XL": 298.98,
   "UD-Q4_K_XL": 379.8,
   "UD-Q5_K_XL": 448.81,
   "UD-Q6_K_XL": 534.56,
   "UD-Q8_K_XL": 726.65
  }
 },
 "runsWith": [
  "vLLM",
  "SGLang",
  "llama.cpp (Ollama, LM Studio)",
  "MLX (Mac)"
 ],
 "verified": null,
 "updated": "2026-11-06T21:02:26.190Z"
}