NVIDIA, Arm and Intel present FP8 Formats for Deep Learning
Presented FP8 Formats for Deep Learning — proposed two encodings, E4M3/E5M2, and claimed performance equal to 16-bit on a 175B language model
Why it matters
As a joint proposal by the three companies, it made FP8 the de facto standard format. A specification agreement that prevented format fragmentation
Source — Primary sources first; secondary reports are labelled “Press”
Paperarxiv.orghttps://arxiv.org/abs/2209.05433Added in Week 35, 2026 · See that week