blocks|key|1480577|text|在NVidia的网站上有类似的quite+a+few+samples。请记住，某些事情，如排序，需要特殊的算法来实现有效的并行性，并且可能不如单核上的非线程算法那么有效。|type|unstyled|depth|inlineStyleRanges|entityRanges|offset|length|data|1480578|entityMap|0|LINK|mutability|MUTABLE|url|http://www.nvidia.com/object/cuda_get_samples.html^0|F|J|0|0^^$0|@$1|2|3|4|5|6|7|L|8|@]|9|@$A|M|B|N|1|O]]|C|$]]|$1|D|3|-4|5|6|7|P|8|@]|9|@]|C|$]]]|E|$F|$5|G|H|I|C|$J|K]]]]

There are <a href="http://www.nvidia.com/object/cuda_get_samples.html" rel="nofollow noreferrer">quite a few samples</a> of this sort of thing on NVidia's website. Bear in mind that some things such as sorting need special algorithms for efficient parallelism and may not be quite as efficient as a non-threaded algorithm on a single core.

blocks|key|1603520|text|GPU是高度专业化的硬件，旨在非常好地完成一小部分任务，并且高度并行。这基本上是算术运算(特别是单精度浮点数学，尽管较新的GPU在双精度下做得很好)。因此，它们只适用于特定的算法。我不确定排序是否适合这个类别(至少在一般情况下是这样)。|type|unstyled|depth|inlineStyleRanges|entityRanges|data|1603521|更常见的例子是金融工具的定价，大量的矩阵数学，甚至defeating+encryption+(通过蛮力)。话虽如此，我还是找到了Fast+parallel+GPU-sorting+using+a+hybrid+algorithm。|offset|length|1603522|另一个经常引用的例子是running+SETI@HOME+on+an+Nvidia+GPU，但它是将苹果与橙子进行比较。与CPU通常所做的相比，GPU的工作单元是不同的(并且非常有限)。|1603523|entityMap|0|LINK|mutability|MUTABLE|url|http://securityandthe.net/2008/10/12/russian-researchers-achieve-100-fold-increase-in-wpa2-cracking-speed/|1|http://portal.acm.org/citation.cfm?id=1412828|2|http://setiathome.berkeley.edu/cuda.php^0|0|P|K|0|1S|1E|1|0|B|Y|2|0^^$0|@$1|2|3|4|5|6|7|T|8|@]|9|@]|A|$]]|$1|B|3|C|5|6|7|U|8|@]|9|@$D|V|E|W|1|X]|$D|Y|E|Z|1|10]]|A|$]]|$1|F|3|G|5|6|7|11|8|@]|9|@$D|12|E|13|1|14]]|A|$]]|$1|H|3|-4|5|6|7|15|8|@]|9|@]|A|$]]]|I|$J|$5|K|L|M|A|$N|O]]|P|$5|K|L|M|A|$N|Q]]|R|$5|K|L|M|A|$N|S]]]]

GPUs are highly specialized hardware designed to do a small set of tasks very well and highly parallelized. This is basically arithmetic (particularly single precision floating point math although newer GPUs do quite well with double precision). As such they're only suited to particular algorithms. I'm not sure if sorting fits that category (in the general case at least).

More common examples are pricing of financial instruments, large amounts of matrix maths and even <a href="http://securityandthe.net/2008/10/12/russian-researchers-achieve-100-fold-increase-in-wpa2-cracking-speed/" rel="noreferrer">defeating encryption</a> (by brute force). That being said, I did find <a href="http://portal.acm.org/citation.cfm?id=1412828" rel="noreferrer">Fast parallel GPU-sorting using a hybrid algorithm</a>.

Another commonly quoted example is <a href="http://setiathome.berkeley.edu/cuda.php" rel="noreferrer">running SETI@HOME on an Nvidia GPU</a> but it's comparing apples to oranges. The units of work for GPUs are different (and highly limited) compared to what CPUs ordinarily do.

blocks|key|1605359|text|看一看|type|unstyled|depth|inlineStyleRanges|entityRanges|data|1605360|CUDA是一个并行算法的CUDA库，其接口类似于C%2B%2B标准模板库(+|1605361|+)。Thrust为图形处理器编程提供了一个灵活的高级接口，极大地提高了开发商的生产力。|blockquote|1605362|1605363|entityMap^0|0|0|0|0^^$0|@$1|2|3|4|5|6|7|J|8|@]|9|@]|A|$]]|$1|B|3|C|5|6|7|K|8|@]|9|@]|A|$]]|$1|D|3|E|5|F|7|L|8|@]|9|@]|A|$]]|$1|G|3|-4|5|6|7|M|8|@]|9|@]|A|$]]|$1|H|3|-4|5|6|7|N|8|@]|9|@]|A|$]]]|I|$]]

Have a look at <a href="http://code.google.com/p/thrust/" rel="noreferrer">thrust</a>:

<blockquote>
 Thrust is a CUDA library of parallel
 algorithms with an interface
 resembling the C++ Standard Template
 Library (STL). Thrust provides a
 flexible high-level interface for GPU
 programming that greatly enhances
 developer productivity.
</blockquote>

blocks|key|1603564|text|对GPGPU引用的任何性能数据都要非常小心。许多人喜欢发布真正令人印象深刻的数字，这些数字没有考虑到从中央处理器到图形处理器的输入数据和输出数据返回所需的传输时间，这两者都越过了PCIe瓶颈。|type|unstyled|depth|inlineStyleRanges|entityRanges|data|1603565|entityMap^0|0^^$0|@$1|2|3|4|5|6|7|D|8|@]|9|@]|A|$]]|$1|B|3|-4|5|6|7|E|8|@]|9|@]|A|$]]]|C|$]]

BE WARY, VERY WARY of any performance numbers quoted for GPGPU. Lots of people like to post really impressive numbers that don't take into consideration the transfer time needed to get the input data from the CPU to the GPU and the output data back, both going over a PCIe bottleneck.

blocks|key|303335|text|图片大小调整在许多接受图片上传的网站上都是很常见的。|type|unstyled|depth|inlineStyleRanges|entityRanges|data|303336|用绝对最低质量选项和最近邻采样在C#中调整2600ish+x+2000ish+2MB+jpeg图像(到512x512)的大小需要23.5毫秒。使用的函数是基于graphics.DrawImage()的函数。CPU使用率也为%2521.5。|offset|length|style|CODE|303337|在GPU端提取"rgba+byte+array“并将其发送到C#，然后在GPU中调整大小并将结果返回到图像中花费了6.3毫秒，CPU使用率为%2512.7。这是用只有320核的便宜了55%25的gpu完成的。|303338|只有3.73倍的加速倍数。|303339|这里的限制因素是，发送提取的20MB+rgb数据(jpeg只有2MB!)转到GPU。这个耗时的部分几乎占总时间的%2590，包括C#端字节数组提取！所以我猜，如果提取部分也能在GPU上完成，至少会有30倍的加速。|303340|30倍还不错。|303341|然后，您可以使用调整大小的层来管道提取层，以隐藏内存复制延迟，从而获得更快的速度！这可能是40X-50X。|303342|然后增加采样的质量(例如双三次而不是最近邻)，你在GPU方面就更有优势了。添加5x5高斯滤波器仅增加了0.77毫秒。在此基础上，CPU将获得更高的时间，特别是如果所需的高斯参数不同于C#.Net实现。|303343|303344|即使你对加速比不满意，卸载到GPU上并在CPU上有一个“空闲的核心”对于将更多的工作推到该服务器上仍然是有利的。|303345|现在加上GPU功耗水平(在本例中为30W与125W+)，它就更有优势了。|303346|303347|CPU很难在|303348|+C[i]=A[i]%2BB[i]|code-block|syntax|javascript|303349|当两端都在优化的代码上运行时，你仍然可以将一半的数组卸载到图形处理器上，同时使用CPU%2BGPU更快地完成基准测试。|303350|303351|GPU不是为非均匀的工作而构建的。GPU有很深的流水线，因此在停顿后由于分支而站立起来，需要太长的时间。同样，SIMD类型的硬件会强制它对其上的所有工作项执行相同的操作。当工作项做与组不同的事情时，它会丢失跟踪并在整个SIMD流水线中添加气泡，或者只是其他工作项等待同步点。因此，清理同时影响深度和广度流水线区域，并使其在完全混乱的条件下甚至比CPU更慢。|303352|entityMap^0|0|27|K|0|0|0|0|0|0|0|0|0|0|0|0|0|0|0|0^^$0|@$1|2|3|4|5|6|7|1D|8|@]|9|@]|A|$]]|$1|B|3|C|5|6|7|1E|8|@$D|1F|E|1G|F|G]]|9|@]|A|$]]|$1|H|3|I|5|6|7|1H|8|@]|9|@]|A|$]]|$1|J|3|K|5|6|7|1I|8|@]|9|@]|A|$]]|$1|L|3|M|5|6|7|1J|8|@]|9|@]|A|$]]|$1|N|3|O|5|6|7|1K|8|@]|9|@]|A|$]]|$1|P|3|Q|5|6|7|1L|8|@]|9|@]|A|$]]|$1|R|3|S|5|6|7|1M|8|@]|9|@]|A|$]]|$1|T|3|-4|5|6|7|1N|8|@]|9|@]|A|$]]|$1|U|3|V|5|6|7|1O|8|@]|9|@]|A|$]]|$1|W|3|X|5|6|7|1P|8|@]|9|@]|A|$]]|$1|Y|3|-4|5|6|7|1Q|8|@]|9|@]|A|$]]|$1|Z|3|10|5|6|7|1R|8|@]|9|@]|A|$]]|$1|11|3|12|5|13|7|1S|8|@]|9|@]|A|$14|15]]|$1|16|3|17|5|6|7|1T|8|@]|9|@]|A|$]]|$1|18|3|-4|5|6|7|1U|8|@]|9|@]|A|$]]|$1|19|3|1A|5|6|7|1V|8|@]|9|@]|A|$]]|$1|1B|3|-4|5|6|7|1W|8|@]|9|@]|A|$]]]|1C|$]]

Image resizing must be common on many websites that accept image uploads.

Resizing a 2600ish x 2000ish 2MB jpeg image (to 512x512) took 23.5 milliseconds in C# with absolute lowest quality options and nearest neighbour sampling. Used function was <code>graphics.DrawImage()</code> based one. CPU usage was also %21.5.

Getting "rgba byte array" extraction on C# side and sending it to GPU and resizing in GPU and getting results back into an image took 6.3 milliseconds and CPU usage was %12.7. This was done with a %55 cheaper gpu with just 320 cores.

Only 3.73X speedup multiplier.

The limiting factor here was, sending the extracted 20MB rgb data (jpeg is only 2MB!) to GPU. That time consuming part was nearly %90 of total time, including C# side byte array extraction! So I gues there would be about 30X speedup at least if extraction part could be done in GPU too.

30X is not bad.

Then you could pipeline the extraction layer with the resizing layer to hide memory copy latency to get even more speed! This could be 40X-50X.

Then increase the quality of sampling(such as bicubic instead of nearest neighbor), you have even more advantage in GPU side. Adding a 5x5 Gaussian filter added only 0.77 milliseonds. CPU would get some higher time on top of that, especially if the Gaussian parameters needed are different than C#.Net implementation.

<hr>

Even if you are not satisfied with speedup ratio, offloading to GPU and having a "free core" on the CPU is still advantageous for pushing more work to that server.

Now add the fact of GPU power consumption levels(30W vs 125W in this example), it is much more advantageous.

<hr>

CPU could hardly win in

<pre><code> C[i]=A[i]+B[i]
</code></pre>

benchmarks when both sides run on optimized codes and you can still offload half of arrays to GPU and finish quicker using CPU+GPU at the same time.

<hr>

GPU is not built for non-uniform works. GPUs have deep pipelines so standing up after a stall because of branching, takes too long. Also SIMD type hardware forces it to do same thing on all workitems on it. When a workitem does a different thing than the group, it loses track and adds bubbles in whole SIMD pipeline or simply others wait for sync point. So brancing affects both deep and wide pipeline areas and make it even slower than CPU in perfectly chaotic conditions.

I'm interested to know if any common algorithms (sorting, searching, graphs, etc.) have been ported to OpenCL (or any GPU language), and how the performance compares to the same algorithm executed by the CPU. I'm specifically interested in the results (numbers).

Thanks!

GPU vs CPU performance for common algorithms

翻译质量差，导致语言生硬或混乱。

没有提供实际的解决方法或示例。

解答不清晰，无法理解或解决问题。

页面排版不美观，阅读体验差。

文章

问答

视频

教程

学习中心

腾讯云实验室

直播

竞赛

腾讯云代码分析专区

腾讯iOA零信任安全管理系统专区

腾讯云架构师技术同盟交流圈

腾讯云数据库专区

腾讯云智能顾问专区

腾讯云原生专区

腾讯混元专区

腾讯云TCE专区

腾讯云Lighthouse专区

腾讯云HAI专区

腾讯云Edgeone专区

腾讯云存储专区

腾讯云智能专区

腾讯轻联专区 

腾讯云开发专区

TAPD专区

腾讯轻量云游戏服专区

腾讯云最具价值专家

腾讯云架构师技术同盟

腾讯云创作之星

腾讯云开发者先锋

腾讯云AI代码助手

云原生构建

TAPD 敏捷项目管理

Cloud Studio

SDK中心

API中心

命令行工具

功能1上新10个字符

功能2描述100个字符功能2描述100个字符功能2描述100个字符功能2描述100个字符功能2描述100个字符功能2描述100个字符功能2描述100个字符功能2描述100个字符功能2描述100个字符。

功能2上新100个字符功能2上新100个字符功能2上新100个字符功能2上新100个字符功能2上新100个字符功能2上新100个字符功能2上新100个字符功能2上新100个字符功能2上新100个字符。

功能5描述100个字符功能5描述100个字符功能5描述100个字符功能5描述100个字符功能5描述100个字符功能5描述100个字符

功能5上新100个字符功能5上新100个字符功能5上新100个字符功能5上新100个字符功能5上新100个字符功能5上新100个字符功能5上新100个字符功能5上新100个字符功能5上新100个字符功能5上新100个字符

功能4上新

文章&问答评论现已支持表情

全新交互，全新视觉，新增快捷键、悬浮工具栏、高亮块等功能并同时优化现有功能，全面提升创作效率和体验

社区富文本编辑器全新改版！诚邀体验～ 

精选全网热门MCP server，让你的AI更好用 🚀

💥开发者 MCP广场重磅上线！

涵盖代码开发、场景应用、自动测试全流程，助你从零构建专属AI助手

一站式MCP教程库，解锁AI应用新玩法

聚焦“写作效率、视觉美观与运行性能”三方面进行全面升级，为您提供更高效、稳定的创作环境

社区富文本&Markdown编辑器全新改版上线，欢迎大家体验!

诚挚邀请您参与本次调研，分享您的真实使用感受与建议。您的反馈至关重要，感谢您的支持与参与！

社区新版编辑器体验调研

我想知道是否有常见的算法(排序、搜索、图表等)。已经被移植到OpenCL (或任何图形处理器语言)，以及性能如何与由中央处理器执行的相同算法进行比较。我对结果(数字)特别感兴趣。谢谢!

问常用算法的GPU与CPU性能对比
EN

回答 5

Stack Overflow用户

Stack Overflow用户

Stack Overflow用户

社区

活动

圈层

关于

腾讯云开发者

热门产品

热门推荐

更多推荐

问常用算法的GPU与CPU性能对比EN

回答 5

Stack Overflow用户

Stack Overflow用户

Stack Overflow用户

社区

活动

圈层

关于

腾讯云开发者

热门产品

热门推荐

更多推荐

问常用算法的GPU与CPU性能对比
EN