简体   繁体   English

Windows 10上的GPU计算群集比Ubuntu 18.04快

[英]GPU compute cluster faster on Windows 10 than Ubuntu 18.04

Info: My windows and Ubuntu system is the same. 信息:我的Windows和Ubuntu系统是相同的。 I just swap out my boot drive using a removable drive bay. 我只是使用可移动驱动器托架换出了引导驱动器。

*Not using Docker *不使用Docker

CPU: i9-7920X (128 GB DDR4) CPU:i9-7920X(128 GB DDR4)

GPUs: 2 Titan Vs and RTX 2070 (Display) GPU:2个Titan Vs和RTX 2070(显示)

Windows 10: Nvidia Driver 416.81 in TCC cluster and Cuda 10 Windows 10:TCC群集和Cuda 10中的Nvidia驱动程序416.81

Ubuntu Desktop 18.04: Nvidia Driver 410.72 (not sure if need to cluster on Linux) and Cuda 10 Ubuntu Desktop 18.04:Nvidia驱动程序410.72(不确定是否需要在Linux上集群)和Cuda 10

Test Software: Running TensorFlow cifar10 esitimator - same training data - (link at bottom) 测试软件:运行TensorFlow cifar10 esitimator-相同的培训数据-(底部链接)

python cifar10_main.py --data-dir=/tmp --job-dir=/tmp/cifar10 --num-gpus=2 --train-steps=10000 python cifar10_main.py --data-dir = / tmp --job-dir = / tmp / cifar10 --num-gpus = 2 --train-steps = 10000

Windows completes in 420 seconds (7.0 minutes) Ubuntu completes in 614 seconds (10.2 minutes) Windows在420秒(7.0分钟)内完成,Ubuntu在614秒(10.2分钟)内完成。

https://github.com/tensorflow/models/tree/master/tutorials/image/cifar10_estimator https://github.com/tensorflow/models/tree/master/tutorials/image/cifar10_estimator

Why is Ubuntu so much slower? 为什么Ubuntu这么慢? Do the GPUs need to be clustered on Linux or are the different drivers (Ubuntu vs. Windows) the issue? GPU是否需要在Linux上集群,还是不同的驱动程序(Ubuntu与Windows)成为问题?

Thanks. 谢谢。

声明:本站的技术帖子网页,遵循CC BY-SA 4.0协议,如果您需要转载,请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.

 
粤ICP备18138465号  © 2020-2024 STACKOOM.COM