TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

Slide algorithm for training deep neural nets faster on CPUs than GPUs

12 点作者 felixr大约 5 年前

1 comment

motbus3大约 5 年前
I feel skeptical as Intel is involved. Not because of money, but Intel has been putting so much effort on some "suboptimal" enhancements that lead to problems as we've been seeing in the past years such as spectre, meltdown, and other newer ones from past weeks, that it is hard to commit that this one hour performance won't go up to 100 hours because of a workaround patch at operating system level. What do you think?