{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-11-09-015",
    "source_story_id": "tmp-brief-fp4-kv-cache",
    "edition_id": "mp-2026-11-09-morning-0122",
    "edition_url": "https://themachinepress.com/edition/2026-11-09",
    "position": 17,
    "story_type": "ticker",
    "section": "chips-infrastructure",
    "editorial_classification": "editorial",
    "headline": "Four-Bit KV Cache Reaches the Tensor Core",
    "slug": "four-bit-kv-cache-reaches-the-tensor-core",
    "dek": "Algorithm-system co-design keeps reported accuracy near BF16 while raising peak serving throughput 2.4 to 3 times on B300 hardware.",
    "summary": "Algorithm-system co-design keeps reported accuracy near BF16 while raising peak serving throughput 2.4 to 3 times on B300 hardware.",
    "body_text": "BitDecoding-FP4 combines error controls for queries, probabilities and keys with fused paged-cache kernels. Across the authors’ models and NVIDIA B300 setup, NVFP4 stayed within 0.5 accuracy points of BF16 KV and delivered 2.4-to-3-times higher peak serving throughput.",
    "why_it_matters": "Algorithm-system co-design keeps reported accuracy near BF16 while raising peak serving throughput 2.4 to 3 times on B300 hardware.",
    "limitations": [],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-11-09-015/four-bit-kv-cache-reaches-the-tensor-core",
    "json_url": "https://themachinepress.com/story/mp-2026-11-09-015.json",
    "first_published_at": "2026-11-09T09:00:00.000-05:00",
    "modified_at": "2026-11-09T09:00:00.000-05:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-11-09-015-001",
        "text": "Algorithm-system co-design keeps reported accuracy near BF16 while raising peak serving throughput 2.4 to 3 times on B300 hardware.",
        "source_ids": [
          "source-2026-11-09-017"
        ],
        "qualification": null
      }
    ],
    "source_ids": [
      "source-2026-11-09-017"
    ],
    "tags": [
      "KV cache",
      "FP4",
      "LLM serving"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-11-09-017",
      "title": "arXiv preprint 2611.09825, original v1",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2611.09825v1",
      "canonical_url": "https://arxiv.org/abs/2611.09825v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-11-06T13:13:54.000-05:00",
      "accessed_at": "2026-11-09T08:25:54.496-05:00",
      "supports_claim_ids": [
        "claim-mp-2026-11-09-015-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "Four-Bit KV Cache Reaches the Tensor Core",
    "publisher": "The Machine Press",
    "published_at": "2026-11-09T09:00:00.000-05:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-11-09-015/four-bit-kv-cache-reaches-the-tensor-core"
  }
}