NVIDIA introduces cuDNN
Presented cuDNN — proposed an optimized library of deep learning primitives in a role equivalent to BLAS, and reported a 36% performance gain when integrated into Caffe
Why it matters
The first layer that spared frameworks from writing their own kernels
Source — Primary sources first; secondary reports are labelled “Press”
Paperarxiv.orghttps://arxiv.org/abs/1410.0759Added in Week 35, 2026 · See that week