zixuanweeei edited a comment on issue #15621: [WIP] MKL-DNN LBR-GRU Inference Integration (FP32 LBR-GRU) URL: https://github.com/apache/incubator-mxnet/pull/15621#issuecomment-516216911 Performance of the latest commit. @ciyongch I have checked the performance again. Their perf are similar, same as we discussed last time. <table border="0"> <td rowspan=2 height=39 class=xl92 width=60 style='height:29.4pt;width:45pt' align=center>Mode</td> <td rowspan=2 class=xl92 width=39 style='width:29pt' align=center>Layer</td> <td rowspan=2 class=xl92 width=61 style='width:46pt' align=center>Direction</td> <td colspan=2 class=xl91 width=241 style='border-left:none;width:181pt' align=center> a26af2b </td> <td colspan=2 class=xl91 width=241 style='border-left:none;width:181pt' align=center>This PR ( cfc6910 )</td> <td colspan=2 class=xl90 width=127 style='border-left:none;width:96pt' align=center>Gap</td> </tr> <tr> <td>Throughput (samples/sec)</td> <td>Latency (ms)</td> <td>Throughput (samples/sec)</td> <td>Latency (ms)</td> <td>Throughput</td> <td>Latency</td> </tr> <tr> <td>lstm</td> <td>1</td> <td>1</td> <td class=xl75 >630.78</td> <td class=xl76 >4.82</td> <td class=xl75 >670.23</td> <td class=xl76 >4.87</td> <td class=xl94 >1.06</td> <td class=xl95 >0.99</td> </tr> <tr> <td height=20 class=xl77 style='height:15.0pt;border-top:none'>lstm</td> <td class=xl67>1</td> <td class=xl67>2</td> <td class=xl68>313.71</td> <td class=xl69>9.68</td> <td class=xl68>338.51</td> <td class=xl69>9.72</td> <td class=xl96>1.08</td> <td class=xl95 >1.00</td> </tr> <tr> <td height=20 class=xl77 style='height:15.0pt;border-top:none'>lstm</td> <td class=xl67>5</td> <td class=xl67>1</td> <td class=xl68>139.85</td> <td class=xl69>23.59</td> <td class=xl68>138.22</td> <td class=xl69>23.48</td> <td class=xl96>0.99</td> <td class=xl95 >1.00</td> </tr> <tr> <td height=20 class=xl77 style='height:15.0pt;border-top:none'>lstm</td> <td class=xl67>5</td> <td class=xl67>2</td> <td class=xl68>54.63</td> <td class=xl69>51.19</td> <td class=xl68>54.27</td> <td class=xl69>51.28</td> <td class=xl96>0.99</td> <td class=xl95 >1.00</td> </tr> <tr> <td height=20 class=xl73 style='height:15.0pt'>rnn_tanh</td> <td class=xl74 >1</td> <td class=xl74 >1</td> <td class=xl75 >1573.45</td> <td class=xl76 >2.44</td> <td class=xl75 >1576.23</td> <td class=xl76 >2.51</td> <td class=xl94 >1.00</td> <td class=xl95 >0.97</td> </tr> <tr> <td height=20 class=xl77 style='height:15.0pt;border-top:none'>rnn_tanh</td> <td class=xl67>1</td> <td class=xl67>2</td> <td class=xl68>836.43</td> <td class=xl69>4.63</td> <td class=xl68>830.33</td> <td class=xl69>4.67</td> <td class=xl96>0.99</td> <td class=xl95 >0.99</td> </tr> <tr> <td height=20 class=xl77 style='height:15.0pt;border-top:none'>rnn_tanh</td> <td class=xl67>5</td> <td class=xl67>1</td> <td class=xl68>381.32</td> <td class=xl69>11.44</td> <td class=xl68>379.88</td> <td class=xl69>11.50</td> <td class=xl96>1.00</td> <td class=xl95 >1.00</td> </tr> <tr> <td height=20 class=xl77 style='height:15.0pt;border-top:none'>rnn_tanh</td> <td class=xl67>5</td> <td class=xl67>2</td> <td class=xl68>159.76</td> <td class=xl69>24.92</td> <td class=xl68>149.86</td> <td class=xl69>24.90</td> <td class=xl96>0.94</td> <td class=xl95 >1.00</td> </tr> <tr> <td height=20 class=xl73 style='height:15.0pt'>rnn_relu</td> <td class=xl74 >1</td> <td class=xl74 >1</td> <td class=xl75 >1536.55</td> <td class=xl76 >2.65</td> <td class=xl75 >1540.29</td> <td class=xl76 >2.75</td> <td class=xl94 >1.00</td> <td class=xl95 >0.96</td> </tr> <tr> <td height=20 class=xl77 style='height:15.0pt;border-top:none'>rnn_relu</td> <td class=xl67>1</td> <td class=xl67>2</td> <td class=xl68>805.00</td> <td class=xl69>5.09</td> <td class=xl68>807.68</td> <td class=xl69>5.06</td> <td class=xl96>1.00</td> <td class=xl95 >1.01</td> </tr> <tr> <td height=20 class=xl77 style='height:15.0pt;border-top:none'>rnn_relu</td> <td class=xl67>5</td> <td class=xl67>1</td> <td class=xl68>373.27</td> <td class=xl69>12.41</td> <td class=xl68>377.79</td> <td class=xl69>12.32</td> <td class=xl96>1.01</td> <td class=xl95 >1.01</td> </tr> <tr height=19 style='height:14.4pt'> <td height=19 class=xl77 style='height:14.4pt;border-top:none'>rnn_relu</td> <td class=xl67>5</td> <td class=xl67>2</td> <td class=xl68>154.21</td> <td class=xl69>26.93</td> <td class=xl68>153.80</td> <td class=xl69>26.61</td> <td class=xl96>1.00</td> <td class=xl95 >1.01</td> </tr> </table>
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to 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