Skip to Main content Skip to Navigation
Poster communications

A new convolutions algorithm to leverage tensor cores

Abstract : We designed a new algorithm for 2D-convolutions that uses tensor cores. Contrary to cuDNN implementation, it does not rely on batched kernels for efficiency. Results are already better than current libraries in some contexts and could be further optimized.
Document type :
Poster communications
Complete list of metadata

https://hal.archives-ouvertes.fr/hal-02605077
Contributor : Nicolas Gac <>
Submitted on : Saturday, May 16, 2020 - 10:27:21 AM
Last modification on : Thursday, May 27, 2021 - 10:34:41 AM

File

Algorithms_Numerical_Technique...
Files produced by the author(s)

Identifiers

  • HAL Id : hal-02605077, version 1

Citation

Mickael Seznec, Nicolas Gac, F. Orieux, A. Sashala Naik. A new convolutions algorithm to leverage tensor cores. GPU Technology Conference (GTC), May 2020, Silicon Valley, United States. ⟨hal-02605077⟩

Share

Metrics

Record views

105

Files downloads

64