报错`RuntimeError: CUDA out of memory. Tried to allocate 256.00 MiB (GPU 0； 9.78 GiB total capaci

小羊星球

已于 2022-11-16 10:03:49 修改

阅读量4.4k

点赞数 1

分类专栏：计算机视觉文章标签：深度学习 tensorflow cnn

于 2022-09-16 16:20:20 首次发布

本文链接：https://blog.csdn.net/everyxing1007/article/details/126892079

版权

计算机视觉专栏收录该内容

12 篇文章 1 订阅

订阅专栏

报错RuntimeError: CUDA out of memory. Tried to allocate 256.00 MiB (GPU 0; 9.78 GiB total capacity; 8.10 GiB already allocated;
用的3080显卡，内存1MiB / 10018MiB
batch_size只能设置较小的值。

1 和网络结构有关

1）9个卷积块（每块包括两个卷积层），batch_size最大为1；

  self.Conv1 = conv_block(ch_in=3, ch_out=64)
  self.Conv2 = conv_block(ch_in=64, ch_out=128)
  self.Conv3 = conv_block(ch_in=128, ch_out=256)
  self.Conv4 = conv_block(ch_in=256, ch_out=512)
  self.Conv5 = conv_block(ch_in=512, ch_out=512)
  self.Conv6 = conv_block(ch_in=512, ch_out=256)
  self.Conv7 = conv_block(ch_in=256, ch_out=128)
  self.Conv8 = conv_block(ch_in=128, ch_out=64)
  self.Conv9 = conv_block(ch_in=64, ch_out=3)

2） 4个卷积块（每块包括两个卷积层），batch_size最大为8；

 self.Conv1 = conv_block(ch_in=3, ch_out=64)
 self.Conv2 = conv_block(ch_in=64, ch_out=128)
 self.Conv3 = conv_block(ch_in=128, ch_out=64)
 self.Conv4 = conv_block(ch_in=64, ch_out=3)

2 和通道数有关：

1）4个卷积块，通道数3-64-64-64-3，batch_size最大为8；

 self.Conv1 = conv_block(ch_in=3, ch_out=64)
 self.Conv2 = conv_block(ch_in=64, ch_out=64)
 self.Conv3 = conv_block(ch_in=64, ch_out=64)
 self.Conv4 = conv_block(ch_in=64, ch_out=3)

2）4个卷积块，通道数3-64-256-512-3，batch_size最大为2；

 self.Conv1 = conv_block(ch_in=3, ch_out=64)
 self.Conv2 = conv_block(ch_in=64, ch_out=256)
 self.Conv3 = conv_block(ch_in=256, ch_out=512)
 self.Conv4 = conv_block(ch_in=512, ch_out=3)