A new convolutions algorithm to leverage tensor cores
Résumé
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.
Fichier principal
Algorithms_Numerical_Techniques_01_P21960_Mickaël_Seznec_Web.pdf (498.84 Ko)
Télécharger le fichier
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|