← Agent Data Hub · Search endpoints · /llms.txt

LLM Local Inference GPU VRAM Sizing Calculator (FP16, Q8, Q4_K_M)

Calculates GPU VRAM required to load and serve open-weights LLMs (7B, 8B, 14B, 32B, 70B, 405B parameters) plus KV-cache overhead.

Call it (no API key)

GET https://api.agentdatahub.de/api/v1/call/dev-gpu-vram-llm-calculator?params_billions=70&compact=1

Full metadata: without compact=1 · Markdown: format=llm · Only some fields: add &fields=a,b.c · MCP: tool call_api with {"apiId":"dev-gpu-vram-llm-calculator"}

Parameters

Example response (live, compact)

{
  "apiId": "dev-gpu-vram-llm-calculator",
  "data_freshness": "cached",
  "data": {
    "model_parameters_billions": 70,
    "vram_required_with_kv_cache_gb": {
      "fp16_bf16_gb": 168,
      "int8_q8_gb": 84,
      "int4_q4_k_m_gb": 48.7
    }
  },
  "source": "100% Platform AI Hardware Sizing Engine",
  "license": "100% Platform Owned Compute"
}

Source & license