[英]Deep Learning on GPU
A DGX-1 has quite a lot of power. DGX-1具有相当大的功能。 However, when using it I only utilize 34% of one of eight cards.
但是,使用它时,我仅使用八张卡之一中的34%。 Are there some points and tweaks I might missed?
我可能会错过一些要点和调整吗? I know I can't parallelize everything, there is an upper limit.
我知道我无法并行化所有内容,有一个上限。 But to be honest, I expected a better performance.
但老实说,我希望有更好的表现。
I'm running a nvidia-docker with tensorflow preinstalled. 我正在运行预装了tensorflow的nvidia-docker。 The script running is from dennybritz which is quite well used.
运行的脚本来自dennybritz,使用非常好。 I ran the docker one one card.
我只给码头工人一张卡。 Since it didn't utilized the whole card I didn't gave him second one.
由于它没有利用整张卡,所以我没有给他第二张卡。 Would this have any benefit?
这会有什么好处吗? Of course I could ran multiple instances and pick the best one.
当然,我可以运行多个实例并选择最佳实例。 But I'd rather have results sooner when having 170TFLOPS accessible.
但是当我可以访问170TFLOPS时,我宁愿早日得到结果。
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.