kaggle notebook的配置查看以及kaggle的GPU与本地显卡性能比较

import torch
print(torch.version.cuda)

!nvcc -V

10.0.130

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2018 NVIDIA Corporation
Built on Sat_Aug_25_21:08:01_CDT_2018
Cuda compilation tools, release 10.0, V10.0.130
###############################查看CPU##############################

!cat /proc/cpuinfo | grep "cpu cores" | uniq

!cat /proc/cpuinfo |grep "processor"|wc -l

运行后会发现是单核双线程

###############################查看GPU##############################

命令:

!apt install lshw -y

!lshw -C display

结果:

  *-display                 
       description: 3D controller
       product: GP100GL
       vendor: NVIDIA Corporation
       physical id: 4
       bus info: pci@0000:00:04.0
       version: a1
       width: 64 bits
       clock: 33MHz
       capabilities: bus_master cap_list
       configuration: driver=nvidia latency=0
       resources: iomemory:80-7f iomemory:c0-bf irq:33 memory:fc000000-fcffffff memory:800000000-bffffffff memory:c00000000-c01ffffff ioport:c000(size=128)

------------------------------------------------------------------------------------

命令:

!nvidia-smi 

结果:

Fri Sep 20 09:32:32 2019       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 418.67       Driver Version: 418.67       CUDA Version: 10.1     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  Tesla P100-PCIE...  On   | 00000000:00:04.0 Off |                    0 |
| N/A   34C    P0    25W / 250W |      0MiB / 16280MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+

 

------------------------------------------------------------------------------------

命令:

!nvidia-smi --query-gpu=name --format=csv,noheader

结果:

Tesla P100-PCIE-16GB

 

 

#################################################################################

根据[1]:最好的经验法则是：如果用RNN，请看带宽；如果用卷积，

请看FLOPS(floating point operation)；如果有钱，上Tensor Cores（除非你必须购买Tesla）

显存爆炸可以调整batch_size

使用了kaggle的Tesla P100-PCIE-16GB和实地Titan 2080做比较．

对RSNA颅内出血比赛的数据集进行测试．一个epoch:

Tesla P100-PCIE-16GB:7小时

Titan 2080：3.5小时不到

爆内存的话可以考虑batch_size

 

另外：

Titan2080有四个接口可以链接四个显示器，但是被训练的显卡连接的显示器可能会发生屏幕抖动。

Reference:

[1]http://m.elecfans.com/article/737945.html