You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@mxnet.apache.org by GitBox <gi...@apache.org> on 2020/09/29 01:40:29 UTC

[GitHub] [incubator-mxnet] samskalicky edited a comment on issue #19247: asnumpy() on cpu is 80 times slower than gpu

samskalicky edited a comment on issue #19247:
URL: https://github.com/apache/incubator-mxnet/issues/19247#issuecomment-700373245


   Cleaned up the code a bit:
   ```python
   import mxnet as mx
   import numpy as np
   import gluoncv 
   import time
   
   # model = self.load_model(best=True)
   model = gluoncv.model_zoo.get_model('faster_rcnn_resnet50_v1b_voc',pretrained=True)
   
   img = mx.random.randn(1,3,608,608,ctx = mx.gpu())
   model.collect_params().reset_ctx(mx.gpu())
   model.hybridize(static_shape=True, static_alloc=True)
   
   #warmup
   print('gpu warmup')
   for i in range(10):
       ids,scores,boxes = model(img)
       ids,scores,boxes = ids.asnumpy(),scores.asnumpy(),boxes.asnumpy()
       mx.nd.waitall()
   
   #gpu test    
   for i in range(10):
       start_time5 = time.time()
       ids,scores,boxes = model(img)
       ids,scores,boxes = ids.asnumpy(),scores.asnumpy(),boxes.asnumpy()
       mx.nd.waitall()
       print('gpu: %4.3f seconds' % (time.time()-start_time5))
   
   model.collect_params().reset_ctx(mx.cpu())
   img = img.as_in_context(mx.cpu())
   
   #warmup
   print('cpu warmup')
   for i in range(10):
       ids,scores,boxes = model(img)
       ids,scores,boxes = ids.asnumpy(),scores.asnumpy(),boxes.asnumpy()
       mx.nd.waitall()
   
   #cpu test
   for i in range(10):
       start_time5 = time.time()
       ids,scores,boxes = model(img)
       ids,scores,boxes = ids.asnumpy(),scores.asnumpy(),boxes.asnumpy()
       mx.nd.waitall()
       print('cpu: %4.3f seconds' % (time.time()-start_time5))
   ```
   Heres the output I get on P3.2xl with mxnet-cu102 (version 1.7.0):
   ```
   gpu warmup
   [01:37:43] src/operator/nn/./cudnn/./cudnn_algoreg-inl.h:97: Running performance tests to find the best convolution algorithm, this can take a while... (set the environment variable MXNET_CUDNN_AUTOTUNE_DEFAULT to 0 to disable)
   gpu: 0.052 seconds
   gpu: 0.052 seconds
   gpu: 0.052 seconds
   gpu: 0.052 seconds
   gpu: 0.052 seconds
   gpu: 0.052 seconds
   gpu: 0.052 seconds
   gpu: 0.052 seconds
   gpu: 0.052 seconds
   gpu: 0.052 seconds
   cpu warmup
   cpu: 2.827 seconds
   cpu: 2.837 seconds
   cpu: 2.829 seconds
   cpu: 2.826 seconds
   cpu: 2.828 seconds
   cpu: 2.830 seconds
   cpu: 2.826 seconds
   cpu: 2.830 seconds
   cpu: 2.832 seconds
   cpu: 2.824 seconds
   ```


----------------------------------------------------------------
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:
users@infra.apache.org



---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@mxnet.apache.org
For additional commands, e-mail: issues-help@mxnet.apache.org