pengzhao-intel commented on issue #8302: Refactor operators & MKLDNN URL: https://github.com/apache/incubator-mxnet/pull/8302#issuecomment-358939263 Regarding the inference performance, the new MKL-DNN shows better performance than MKLML engine as below table. *I will update the training performance data later* **updated 1/21 for more batchsize of the inference** Tested on two sockets SKX-8180 machine (total 56 physical cores): topo (images/sec) | Batchsize |Refactor PR | MKL2017 -- | -- |--| -- Alexnet (224*224 ) |1| 366 | 343 | |32| 1460 | 696 | |64| 1690 | 751 | |128| 1761 | 761 Inception-v3 (299*299 ) |1| 53 | 33 | |32| 214 | 125 | |64| 206 | 125 | |128| 215 | 126 Vgg-16 (224*224 ) |1| 101 | 58 | |32| 145 | 87 | |64| 138 | 87 | |128| 138 | 87 ResNet50 (224*224 )|1 | 80 | 56 | |32| 202 | 166 | |64| 203 | 155 | |128| 194 | 151
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
