�组保留的标签 余下为需要删除的标签
                unset($oldtag[$key]);
            }
        }
    }
    

    if (!empty($oldtag)) {
        $tagids = array();
        foreach ($oldtag as $tagid => $tagname) {
            $tagids[] = $tagid;
        }

        well_oldtag_delete($tagids, $tid);
    }

    $r = well_tag_process($tid, $fid, $create_tag, $tagarr);

    

    return $r;
}

// 删除标签和绑定的主题
function well_oldtag_delete($tagids, $tid)
{
    

    $pagesize = count($tagids);
    $arrlist = well_tag_find_by_tagids($tagids, 1, $pagesize);

    

    $delete_tagids = array(); // 删除
    $tagids = array();
    $n = 0;
    foreach ($arrlist as $val) {
        ++$n;
        
        if (1 == $val['count']) {
            // 只有一个主题
            $delete_tagids[] = $val['tagid'];
            
        } else {
            $tagids[] = $val['tagid'];
        }
        
    }

    

    !empty($delete_tagids) and well_tag_delete($delete_tagids);

    $arlist = well_tag_thread_find_by_tid($tid, 1, $n);
    if ($arlist) {
        $ids = array();
        foreach ($arlist as $val) $ids[] = $val['id'];

        well_tag_thread_delete($ids);
    }

    !empty($tagids) and well_tag_update($tagids, array('count-' => 1));

    
}

// 标签数据处理 $arr=新提交的数组 $tagarr=保留的旧标签
function well_tag_process($tid, $fid, $new_tags = array(), $tagarr = array())
{
    if (empty($tid)) return '';

    

    // 新标签处理入库
    if ($new_tags) {

        

        $threadarr = array();
        $tagids = array();
        $i = 0;
        $size = 5;
        $n = count($tagarr);
        $n = $n > $size ? $size : $size - $n;

        

        foreach ($new_tags as $name) {
            ++$i;
            $name = trim($name);
            $name = stripslashes($name);
            $name = strip_tags($name);
            $name = str_replace(array('&nbsp;', '#', "@", "$", "%", "^", '&', '·', '<', '>', '；', '`', '~', '!', '￥', '……', ';', '?', '？', '-', '—', '_', '=', '+', '.', '{', '}', '|', ':', '：', '、', '/', '。', '[', ']', '【', '】', '‘', '	', '    ', '  ', '   ', '    '), '', $name);
            $name = htmlspecialchars($name, ENT_QUOTES);
            if ($name && $i <= $n) {
                
                // 查询标签
                $read = well_tag_read_name($name);
                
                if ($read) {
                    // 存在 count+1
                    $tagids[] = $read['tagid'];
                    
                } else {
                    // 入库
                    $arr = array('name' => $name, 'count' => 1);
                    
                    $tagid = well_tag_create($arr);
                    FALSE === $tagid and message(-1, lang('create_failed'));
                    $read = array('tagid' => $tagid, 'name' => $name);
                    
                }

                $tag_thread = array('tagid' => $read['tagid'], 'tid' => $tid);

                

                $threadarr[] = $tag_thread;

                

                $tagarr[$read['tagid']] = $read['name'];
            }
        }

        

        !empty($threadarr) and tag_thread_big_insert($threadarr);

        !empty($tagids) and well_tag_update($tagids, array('count+' => 1));
    }

    

    $json = empty($tagarr) ? '' : xn_json_encode($tagarr);

    

    return $json;
}




?>   $v = implode(",", $v);
        $temp[] = $v;
    }
    // 去掉重复的字符串,也就是重复的一维数组
    $temp = array_unique($temp);
    // 再将拆开的数组重新组装
    $output = array();
    foreach ($temp as $k => $v) {
        if ($stkeep) $k = $starr[$k];
        if ($ndformat) {
            $temparr = explode(",", $v);
            foreach ($temparr as $ndkey => $ndval) $output[$k][$ndarr[$ndkey]] = $ndval;
        } else $output[$k] = explode(",", $v);
    }
    return $output;
}

// 合并二维数组 如重复 值以第一个数组值为准
function array2_merge($array1, $array2, $key = '')
{
    if (empty($array1) || empty($array2)) return NULL;
    $arr = array();
    foreach ($array1 as $k => $v) {
        isset($v[$key]) ? $arr[$v[$key]] = array_merge($v, $array2[$k]) : $arr[] = array_merge($v, $array2[$k]);
    }
    return $arr;
}

/*
 * 对二维数组排序 两个数组必须有一个相同的键值
 * $array1 需要排序数组
 * $array2 按照该数组key排序
 * */
function array2_sort_key($array1, $array2, $key = '')
{
    if (empty($array1) || empty($array2)) return NULL;
    $arr = array();
    foreach ($array2 as $k => $v) {
        if (isset($v[$key]) && $v[$key] == $array1[$v[$key]][$key]) {
            $arr[$v[$key]] = $array1[$v[$key]];
        } else {
            $arr[] = $v;
        }
    }

    return $arr;
}




?><!DOCTYPE html><html lang="zh-cn"><head><meta http-equiv="Content-Type" content="text/html; charset=utf-8" /><meta http-equiv="X-UA-Compatible" content="IE=Edge,chrome=1" /><meta http-equiv="Cache-Control" content="no-transform" /><meta http-equiv="Cache-Control" content="no-siteapp" /><meta name="viewport"
          content="width=device-width, initial-scale=1.0, minimum-scale=1.0, maximum-scale=1.0, user-scalable=no"><meta name="referrer" content="origin-when-cross-origin"><meta name="applicable-device" content="pc,mobile" /><meta name="MobileOptimized" content="width" /><meta name="HandheldFriendly" content="true" /><meta name="renderer" content="webkit" /><meta name="keywords" content="训练时GPU不够用的话怎么办办呢" /><meta name="description" content="训练时GPU不够用的话怎么办办呢" /><title>训练时GPU不够用的话怎么办办呢编程频道|福州电脑网
</title><link rel="shortcut icon" href="/view/img/favicon.ico" /><link rel="icon" sizes="32x32" href="/view/img/favicon.ico"><link rel="Bookmark" href="/view/img/favicon.ico" /><link rel="stylesheet" href="/view/template/mitiqin/css/style.css?2.3.0"><link rel="stylesheet" href="/view/template/mitiqin/css/iconfont.css?2.3.0"><script src="/view/template/mitiqin/js/jquery-2.2.4.min.js?2.3.0" type="text/javascript"></script></head><body><div id="header"><div class="header container"><div class="logo"><a href="/"  title="福州电脑网_福州电脑维修_福州电脑之家_福州iThome"><img src="/view/template/mitiqin/img/logo.png" alt="福州电脑网_福州电脑维修_福州电脑之家_福州iThome"></a></div><div id="monavber" class="nav" data-type="index" data-infoid><ul id="nav" class="navbar"><li class="navbar-item" data-active="fid-0"><a href="/">首页</a></li><li class="navbar-item" fid="5" data-active="fid-5"><a href="/biancheng">编程之家</a></li><li class="navbar-item" fid="4" data-active="fid-4"><a href="/xitong">系统教程</a></li><li class="navbar-item" fid="12" data-active="fid-12"><a href="/shuma">数码周边</a></li><li class="navbar-item" fid="13" data-active="fid-13"><a href="/pc">电脑</a></li></ul></div><div id="mnav"><i class="iconfont icon-app"></i></div><div id="search"><i class="iconfont icon-search"></i></div><div class="search"><form id="form-search" action="/operate/search.html"><input type="text" name="keyword" placeholder="关键词" value=""/><button type="submit" value="搜索"><i class="iconfont icon-search"></i></button><input type="hidden" name="range" value="1"></form></div></div></div>
<div class="fzithome-com breadcrumb2 container-w"><div id="b" class="cl"><div class="fzithome-com fl"><i class="iconfont icon-home"></i><a href="/">首页</a>&nbsp;>&nbsp;
<a href="/biancheng">编程之家</a><a href="/biancheng/1740120265a1596056.html" title="首页">训练时GPU不够用的话怎么办办呢</a></div></div></div><div class="fzithome-com main container-w cl"><div class="fzithome-com mainl"><div class="fzithome-com art-post"><div class="fzithome-com title"><h1>训练时GPU不够用的话怎么办办呢</h1><div class="fzithome-com info"><span><a href="/biancheng"><i class="iconfont icon-crossborder-fill"></i>编程之家</a></span><span><i class="iconfont icon-browse"></i>91</span><span><i class="iconfont icon-message-fill"></i>0</span><span>
                        更新时间：2026-04-03 20:26:57</span></div></div><div class="fzithome-com art-content"><article class="single-post"><div class="fzithome-com entry"><path stroke-linecap="round" d="M5,0 0,2.5 5,5z"  style="-webkit-tap-highlight-color: rgba(0, 0, 0, 0);"></path>
                    </svg>
                    <p>1.首先查看能调用的gpu的情况<br /> python<br /> import tensorflow as tf<br /> tf.test.gpu_device_name()<br /> 2. 查看gpu的资源占用情况<br /> nvidia-smi<br /> 3.使用fuser命令显示所有占用nvidia设备的进程processID。<br /> sudo fuser /dev/nvidia*<br /> 4.杀死所有进程<br /> killall -u 用户名</p> 
<p>PS: 我的机器显示有一个GPU设备被ignore了<br /> 解决方法&#xff1a;<br /> 输入export TF_MIN_GPU_MULTIPROCESSOR_COUNT&#61;5</p> 
<p>Note though that the default is set for a reason – if you enable slower GPU by changing that variable, your program may run slower than it would without any GPU available, because TensorFlow will try to put run everything on that GPU.</p></div></article></div><div class="fzithome-com info"><div><span>本文发布于:2025-02-21，感谢您对本站的认可！</span></div><div><span>本文链接:</span><a href="https://www.fzithome.com/biancheng/1740120265a1596056.html" title="训练时GPU不够用的话怎么办办呢">https://www.fzithome.com/biancheng/1740120265a1596056.html</a></div><div><span>版权声明:本站内容均来自互联网，仅供演示用，请勿用于商业和其他非法用途。如果侵犯了您的权益请与我们联系，我们将在24小时内删除。</span></div></div><span class="tag"><i class="iconfont icon-discount" style="font-size:12px;font-weight:bold;opacity:.7;">本文标签：</i><a href="/tag/71418.html" target="_blank">不够用</a><a href="/tag/68085.html" target="_blank">GPU</a></span></div><div class="fzithome-com post_comments" id="comments"><div id="comt-respond" class="commentpost"><h4>发布评论
<span><a rel="nofollow" id="cancel-reply" href="#comment" style="display:none;"><small>取消回复</small></a></span></h4><form action="/comment/create/1596056.html?safe_token=KuRVBj7BzKdps0Y8Oyah6CoD2vfZ0GR_2FeB3MWdoG9evISyrdF5zpYaW8dtS6wt4mq7A4iCjIujq_2F7rAa5a0qeA_3D_3D" method="post" name="saypl" id="frmSumbit"><input type="hidden" name="doctype" value="1" /><input type="hidden" name="quotepid" value="0" /><div id="comment-tools"><div class="fzithome-com tools_text"><textarea placeholder="请在这里留言..." name="message" id="txaArticle" class="text input-block-level comt-area" cols="50" rows="4" tabindex="5"></textarea></div></div><div class="fzithome-com psumbit"><input name="sumbit" type="submit" tabindex="6" value="发布" class="button" /></div></form></div><div class="fzithome-com commentlist"><div class="fzithome-com comment-tab"><div class="fzithome-com come-comt">评论列表<span id="comment_count">（有<span id="infocommentnumarea" style="color:#c81111">0</span>条评论）</span></div></div><ul class="diy-comment"></ul></div></div></div><div class="fzithome-com mainr"><section id="aside_about" class="widget widget_aside_about sb br mb"><div class="avatar"><img class="img" src="/view/template/mitiqin/img/tx.jpg"
             alt="福州电脑网_福州电脑维修_福州电脑之家_福州iThome"/></div><div class="wrap pd"><p class="title">福州电脑网_福州电脑维修_福州电脑之家_福州iThome</p><p class="info">福州电脑维修网(fzithome.com)专业的电脑维修,笔记本维修,上门维修各种电脑,笔记本,平板等,快速上门.电脑知识频道内容覆盖:计算机资讯,电脑基础应用知识,各种电脑故障维修学习,电脑外设产品维修维护,病毒，软件,硬件,常识.</p><ul class="ul clearfix"></ul></div></section><div class="fzithome-com clear"></div><div class="fzithome-com widgets"><h4 class="bar">相关推荐</h4><div class="fzithome-com hot-post"><ul class="clearfix"><li><a href="/biancheng/1729058811a868168.html"title='深度学习环境搭建（GPU）CUDA安装（完全版）' aria-label='深度学习环境搭建（GPU）CUDA安装（完全版）'><span class="sptit">深度学习环境搭建（GPU）CUDA安装（完全版）</span></a></li><li><a href="/biancheng/1729075677a870232.html"title='ARM GPU mali系列产品规划图或天梯图' aria-label='ARM GPU mali系列产品规划图或天梯图'><span class="sptit">ARM GPU mali系列产品规划图或天梯图</span></a></li><li><a href="/biancheng/1729076236a870301.html"title='大模型必备算力：CPU&amp;amp;GPU天梯图（2023年最新版）' aria-label='大模型必备算力：CPU&amp;amp;GPU天梯图（2023年最新版）'><span class="sptit">大模型必备算力：CPU&amp;amp;GPU天梯图（2023年最新版）</span></a></li><li><a href="/biancheng/1729077043a870402.html"title='GPU必知 | 不为RTX 30设限，这般配置发挥每一分能力' aria-label='GPU必知 | 不为RTX 30设限，这般配置发挥每一分能力'><span class="sptit">GPU必知 | 不为RTX 30设限，这般配置发挥每一分能力</span></a></li><li><a href="/biancheng/1729090420a872081.html"title='CPU GPU天梯图' aria-label='CPU GPU天梯图'><span class="sptit">CPU GPU天梯图</span></a></li><li><a href="/biancheng/1729149945a879713.html"title='tensoflow gpu版安装 win10+anaconda3+python3.6+gtx950m' aria-label='tensoflow gpu版安装 win10+anaconda3+python3.6+gtx950m'><span class="sptit">tensoflow gpu版安装 win10+anaconda3+python3.6+gtx950m</span></a></li><li><a href="/biancheng/1729151146a879912.html"title='Windows 10 、GPU 1080 ti 安装Anaconda， GPU版本的TensorFlow 详细教程' aria-label='Windows 10 、GPU 1080 ti 安装Anaconda， GPU版本的TensorFlow 详细教程'><span class="sptit">Windows 10 、GPU 1080 ti 安装Anaconda， GPU版本的TensorFlow 详细教程</span></a></li><li><a href="/biancheng/1733684400a1141918.html"title='windows系统下实现lammps的GPU加速计算' aria-label='windows系统下实现lammps的GPU加速计算'><span class="sptit">windows系统下实现lammps的GPU加速计算</span></a></li><li><a href="/biancheng/1733902032a1170097.html"title='适用于 LLM 推理的最佳 NVIDIA GPU：综合指南' aria-label='适用于 LLM 推理的最佳 NVIDIA GPU：综合指南'><span class="sptit">适用于 LLM 推理的最佳 NVIDIA GPU：综合指南</span></a></li><li><a href="/biancheng/1738083434a1447471.html"title='电脑硬件性能：HDD + SSD + CPU + GPU' aria-label='电脑硬件性能：HDD + SSD + CPU + GPU'><span class="sptit">电脑硬件性能：HDD + SSD + CPU + GPU</span></a></li><li><a href="/biancheng/1738280419a1473057.html"title='配置GPU环境【Win10基于anaconda虚拟环境】' aria-label='配置GPU环境【Win10基于anaconda虚拟环境】'><span class="sptit">配置GPU环境【Win10基于anaconda虚拟环境】</span></a></li><li><a href="/biancheng/1739669978a1549451.html"title='在Windows 710系统下如何安装TensorFlow GPU版' aria-label='在Windows 710系统下如何安装TensorFlow GPU版'><span class="sptit">在Windows 710系统下如何安装TensorFlow GPU版</span></a></li><li><a href="/biancheng/1742506280a1753819.html"title='【配置GPU运算】win7 64位 + CUDA 10.1 + cuDNN v7.0.5 安装[ 第0步 ] 查看当前GPU所支持的哪个版本的CUDA' aria-label='【配置GPU运算】win7 64位 + CUDA 10.1 + cuDNN v7.0.5 安装[ 第0步 ] 查看当前GPU所支持的哪个版本的CUDA'><span class="sptit">【配置GPU运算】win7 64位 + CUDA 10.1 + cuDNN v7.0.5 安装[ 第0步 ] 查看当前GPU所支持的哪个版本的CUDA</span></a></li><li><a href="/biancheng/1742619089a1768535.html"title='GPU内存不够用怎么办？（5分钟扩充显存到32GB）' aria-label='GPU内存不够用怎么办？（5分钟扩充显存到32GB）'><span class="sptit">GPU内存不够用怎么办？（5分钟扩充显存到32GB）</span></a></li><li><a href="/biancheng/1743890308a1848120.html"title='Windows安装Pytorch-GPU版本' aria-label='Windows安装Pytorch-GPU版本'><span class="sptit">Windows安装Pytorch-GPU版本</span></a></li><li><a href="/biancheng/1754087323a2547295.html"title='虚拟内存要不要删除，电脑c盘不够用？' aria-label='虚拟内存要不要删除，电脑c盘不够用？'><span class="sptit">虚拟内存要不要删除，电脑c盘不够用？</span></a></li><li><a href="/biancheng/1754750306a2639521.html"title='GPU发展史（二）：改变游戏规则的3Dfx Voodoo' aria-label='GPU发展史（二）：改变游戏规则的3Dfx Voodoo'><span class="sptit">GPU发展史（二）：改变游戏规则的3Dfx Voodoo</span></a></li><li><a href="/biancheng/1754771838a2641775.html"title='GPT4All:免费使用本地运行无需GPU的基于C++的GPT模型和UI: A free-to-use, locally running, privacy-aware chatbot. No GPU' aria-label='GPT4All:免费使用本地运行无需GPU的基于C++的GPT模型和UI: A free-to-use, locally running, privacy-aware chatbot. No GPU'><span class="sptit">GPT4All:免费使用本地运行无需GPU的基于C++的GPT模型和UI: A free-to-use, locally running, privacy-aware chatbot. No GPU</span></a></li><li><a href="/biancheng/1755028109a2660397.html"title='openstack 配置GPU直通（pci_passthrough）' aria-label='openstack 配置GPU直通（pci_passthrough）'><span class="sptit">openstack 配置GPU直通（pci_passthrough）</span></a></li><li><a href="/biancheng/1755028742a2660528.html"title='OpenStack下利用透传技术(passthrough)对GPU实现虚拟化' aria-label='OpenStack下利用透传技术(passthrough)对GPU实现虚拟化'><span class="sptit">OpenStack下利用透传技术(passthrough)对GPU实现虚拟化</span></a></li></ul></div></div><div class="fzithome-com clear"></div><div class="widget widget_tags"><h4 class="bar">标签列表</h4><ul><li class="submenu"><a target="_blank" href="/tag/244962.html">对分区进</a></li><li class="submenu"><a target="_blank" href="/tag/244945.html">动更新的</a></li><li class="submenu"><a target="_blank" href="/tag/244932.html">打开已存</a></li><li class="submenu"><a target="_blank" href="/tag/244931.html">子文件</a></li><li class="submenu"><a target="_blank" href="/tag/244920.html">位客户端</a></li><li class="submenu"><a target="_blank" href="/tag/244914.html">还有必要</a></li><li class="submenu"><a target="_blank" href="/tag/244862.html">却截图出</a></li><li class="submenu"><a target="_blank" href="/tag/244834.html">中顿号失</a></li><li class="submenu"><a target="_blank" href="/tag/244805.html">面板添加</a></li><li class="submenu"><a target="_blank" href="/tag/244760.html">该网站的</a></li><li class="submenu"><a target="_blank" href="/tag/244730.html">先级</a></li><li class="submenu"><a target="_blank" href="/tag/244706.html">取字节集</a></li><li class="submenu"><a target="_blank" href="/tag/244700.html">打不开的</a></li><li class="submenu"><a target="_blank" href="/tag/244644.html">网页长截</a></li><li class="submenu"><a target="_blank" href="/tag/244575.html">返回值为</a></li><li class="submenu"><a target="_blank" href="/tag/244561.html">在一些情</a></li><li class="submenu"><a target="_blank" href="/tag/244518.html">点碎片</a></li><li class="submenu"><a target="_blank" href="/tag/244508.html">安全拦截</a></li><li class="submenu"><a target="_blank" href="/tag/244484.html">不要换行</a></li><li class="submenu"><a target="_blank" href="/tag/244468.html">用户的受</a></li></ul></div><div class="fzithome-com clear"></div></div></div>