如何以编程方式为TensorFlow中的所有可用内核构建CUDA JIT高速缓存？ [英] How to build CUDA JIT caches for all available kernels in TensorFlow programmatically?

问题描述

我遇到了此问题中讨论的GTX 1080卡和nvidia-docker的首次运行缓慢问题>。

I encountered the "first-run slow-down" problem with GTX 1080 cards and nvidia-docker as discussed in this question.

我使用的是来自其官方pip包和基于nvidia-docker的Ubuntu 16.04基本映像的自定义docker镜像。

I'm using the TensorFlow build from its official pip package and a custom docker image based on nvidia-docker's Ubuntu 16.04 base image.

如何使TensorFlow在Dockerfile中以编程方式加载（和构建JIT缓存）所有注册的CUDA内核？ （而不是使用 TF_CUDA_COMPUTE_CAPABILITIES 环境变量手动构建TensorFlow）

How do I make TensorFlow to load (and build JIT caches) all registered CUDA kernels programmatically in a Dockerfile? (rather than manually building TensorFlow using TF_CUDA_COMPUTE_CAPABILITIES environment variable)

推荐答案

似乎是不容易的方法来实现这一点，因为CUDA运行时隐式，lazily编译丢失cubin从给定的内核源。

There seems to be no easy way to achieve this since CUDA runtime implicitly, lazily compiles missing cubin from the given kernel sources as discussed here.

通过自己重建TensorFlow解决了这个问题，使用一些帮助脚本来检测当前CUDA / GPU配置并生成所需的TensorFlow配置参数（ detect-cuda.py ， build-tensorflow .sh ）。

Solved this problem by rebuilding TensorFlow by myself, with some helper scripts to detect current CUDA/GPU configs and generate required TensorFlow configuration parameters (detect-cuda.py, build-tensorflow.sh).

这篇关于如何以编程方式为TensorFlow中的所有可用内核构建CUDA JIT高速缓存？的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！