"FP8 is all you need" vs "A naive Ozaki II implementation with r=10 residue planes per warp can spill, raising β above 1 and erasing the projected speedup."
Which one wins?
#HPC
Glenn K. Lockwood
=>
FP8 is All You Need (Part 1): Debunking HW FP64 as the HPC Holy Grail, Satoshi Matsuoka, RIKEN, arXiv, May 28, 2026 arxiv.org/abs/2606.06510
A TME Model & Implementation Strategy for Ozaki Scheme II on Memory-Bound Workloads in the Post-FP64 Era
FugakuNEXT, RIKEN, May 29 bsky.app/profile/ogaw...
OGAWA, Tadashi
=>
Basic Design Technical Report for the New Flagship System "FugakuNEXT", Ver 1.1, RIKEN, May 29, 2026 (275 pp) www.r-ccs.riken.jp/fugaku-next/...
「富岳」の次世代となる新たなフラッグシップシステムにおける基本設計技術報告書、第1.1版、理研、2026年5月29日 (268頁) www.r-ccs.riken.jp/fugaku-next/...
Fine-Grained DRAM, NVIDIA bsky.app/profile/ogaw...
OGAWA, Tadashi
=>
Fine-Grained DRAM, NVIDIA
"Insights From NVIDIA Research", Bill Dally, GTC 2026 bsky.app/profile/ogaw...
Novel DRAM Memory Technology
Mike O'Connor, PhD Thesis, 2021 x.com/ogawa_tter/s...
MICRO 2017
「富岳NEXT」プロジェクト WS、3月6日 bsky.app/profile/ogaw...
NVIDIA GPU
先端技術を採⽤した積層メモリ (Stacked Memory)?
=>
"Floating Point Emulation in NVDIA Math Libraries", Samuel Rodriguez, NGT - Openlab "Optimising Floating Point Precision" WS, Jul 1
(44:07) indico.cern.ch/event/153840...
indico.cern.ch/event/153840...
FP64 with Ozaki-I method will be released 2H/2025
K. Ozaki, Jul 2 bsky.app/profile/ogaw...
OGAWA, Tadashi
=>
"Emulating Matrix Multiplication Using Mixed-Precision Computation", K. Ozaki, NGT - Openlab "Optimising Floating Point Precision" WS, Jul 2
(MP4) indico.cern.ch/event/153840...
indico.cern.ch/event/153840...
Ozaki Scheme II, Apr 27 (10) arxiv.org/abs/2504.08009
Aug 8 (6) bsky.app/profile/ogaw...
OGAWA, Tadashi
=>
"High-Performance and Power-Efficient Emulation of Matrix Multiplication using INT8 Matrix Engines", Y. Uchino, K. Ozaki, T. Imamura, arXiv, Aug 8, 2025 arxiv.org/abs/2508.03984
DGEMM & SGEMM emulation based on Ozaki scheme
II using INT8
Satoshi Matsuoka, RIKEN, Jul 23 bsky.app/profile/ogaw...
OGAWA, Tadashi
=>
"Of Oxen and Chickens: Seymour Cray's Legacy in the Convergent Future of HPC and AI", Satoshi Matsuoka, RIKEN, TPC Seminar Series, Jul 23, 2025 anl.app.box.com/s/05t42yots1...
AI for Science Supercomputer, TRIP-AGIS, Jul 28 bsky.app/profile/ogaw...
Grace Blackwell
15.539EF (FP8)
64.16PF (FP64)
OGAWA, Tadashi
=>
AI for Science開発用スーパーコンピュータのシステムが決定、2025年7月28日 www.riken.jp/pr/news/2025...
TRIP-AGIS 計算基盤開拓プロジェクト
NVIDIA GH200
64.16 PF (FP64)
15.539 EF (FP8)
富士通
泰地 真弘人、TRIP-AGISプログラムディレクター、2024年11月2日 www.youtube.com/watch?v=rzj0...
富岳NEXT、6月18日 bsky.app/profile/ogaw...
MONAKA-X
NVLink Fusion: MONAKA, May 18
=>
NVIDIA Unveils NVLink Fusion for Industry to Build Semi-Custom AI Infrastructure w/ NVIDIA Partner Ecosystem, May 18, 2025 nvidianews.nvidia.com/news/nvidia-...
MediaTek, Marvell, Alchip, ...
Fujitsu & QCOM Each Plan to Build Custom CPUs Coupled w/ NVIDIA GPUs
FUJITSU-MONAKA x.com/ogawa_tter/s...