method · first seen 18 Oct 2026
FP64 emulation on low-precision matrix engines
Also: DGEMM emulation, emulated double-precision GEMM, INT8 emulation of double precision
Computing double-precision matrix multiplication by decomposing it into integer matrix products on faster low-precision hardware units.
Mentions per week
PapersCodeDiscussion
Earliest sources
- Emulating DGEMM and ZGEMM with INT8 Matrix Engines on Intel Arc GPUs: Ozaki Scheme II on the Xe2 ArchitecturePapers · 18 Oct 2026 · Rostislavs Rostovskis, Aleksandrs Platonenko
- Providing Accuracy Guarantees for the Fixed-Point Emulation of Matrix MultiplicationPapers · 13 Nov 2026 · Cole Brower, Peter Caday, John Gunnels
Latest sources
- Providing Accuracy Guarantees for the Fixed-Point Emulation of Matrix MultiplicationPapers · 13 Nov 2026 · Cole Brower, Peter Caday, John Gunnels
- Emulating DGEMM and ZGEMM with INT8 Matrix Engines on Intel Arc GPUs: Ozaki Scheme II on the Xe2 ArchitecturePapers · 18 Oct 2026 · Rostislavs Rostovskis, Aleksandrs Platonenko