Metraipolis › archive › dc_f9f64abbd840

Fused-Softmax for efficient quantized attention — arXiv:2611.09225

Source
arXiv (en)
Original
https://arxiv.org/abs/2611.09225
License (as archived)
CC0-1.0
Archived
2026-11-09T22:33:56.007Z
SHA-256 of the archived text
30c63527271e44c6ccfa8da0aeed5f3038493aa4ad0eaf07b9426bee8807c712
Stable URL of this page
https://metraipolis.com/documento/dc_f9f64abbd840

How to cite

Fused-Softmax for efficient quantized attention — arXiv:2611.09225 — arXiv (en), https://arxiv.org/abs/2611.09225 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_f9f64abbd840 (Metraipolis archive dc_f9f64abbd840, sha256 30c63527271e44c6ccfa8da0aeed5f3038493aa4ad0eaf07b9426bee8807c712)
BibTeX
@misc{dcf9f64abbd840,
  title = {Fused-Softmax for efficient quantized attention — arXiv:2611.09225},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.09225},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_f9f64abbd840; sha256 30c63527271e44c6ccfa8da0aeed5f3038493aa4ad0eaf07b9426bee8807c712},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.09225. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Low-precision matrix throughput on modern GPUs is growing far faster than vector-arithmetic throughput. Attention therefore cannot realize the available GEMM performance because softmax has become the pipeline bottleneck. We introduce Fused-Softmax Attention, a training-free algorithm that reformulates and approximates softmax to fold its dominant operations into the attention GEMMs, moving work from vector units to matrix cores. Implemented for the AMD CDNA 4 architecture, our MXFP8, MXFP6 and MXFP4 kernels reach approximately 3.0, 4.0 and 4.6 PFLOP/s on an AMD Instinct MI355X GPU, yielding speedups of $2.1\times$, $2.8\times$ and $3.2\times$ over the BF16 attention. Importantly, the additional error introduced by Fused-Softmax remains below the error floor set by quantizing the GEMM operands. On Wan2.2-T2V-A14B, a hybrid MXFP6/4 schedule delivers up to $2.05\times$ end-to-end speedup while preserving the visual quality of the outputs. The kernels ship in AMD's open-source AITER library, and they powered AMD's MLPerf Inference v6.1 Wan2.2 submission, which ran $1.7\times$ faster than AMD's previous-round result on the same hardware. Our results show that redesigning softmax to utilize the matrix pipeline enables attention to capitalize on continued growth in low-precision matrix throughput.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Artificial Intelligence (cs.AI) ; Machine Learning (cs.LG); Performance (cs.PF).
Pagina del lavoro: https://arxiv.org/abs/2611.09225 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-11-09T22:30:47.268Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).