用于 MD5 加密的 Nvidia CUDA 或 ATI Stream?
当我看到一些我不知道的有关 ATI 的统计数据时,我正准备购买 Nvidia GTX 470。
ATI 卡上的流是否能更好地执行加密和解密所需的数学运算?
I was just set to purchase a Nvidia GTX 470 when i saw some statistics regarding ATI which i didn't know.
Is stream on an ATI card much better at performing the maths required for encryption and decryption?
如果你对这篇内容有疑问,欢迎到本站社区发帖提问 参与讨论,获取更多帮助,或者扫码二维码加入 Web 技术交流群。
绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱,如果其他用户或者作者回复了您的评论,将不能在第一时间通知您!
发布评论
评论(3)
我没有马参加这次比赛;但作为一个尝试自己进行 GPGPU 开发并且知道许多其他人也在做同样事情的人,让我观察一下:
接近峰值性能
ATI卡更难,因为不
只是你必须打破
计算单元高达
许多计算核心,你实际上
必须进行 4 路矢量化
对众多的每一个进行计算
核心。对于简单的事情,比如
加密或 MD5,这可能是
简单,但这是另一个
您必须经历的步骤。
NVidia SDK 附带的免费开发工具(调试器和分析器,以及整个在线开发社区)比 ATI 附带的任何工具都要有用得多。说 OpenCL 就是 OpenCL 很好,但这并不能帮助您找到偷偷摸摸的错误,也不能帮助您找出为什么您的占用率低于您认为应有的水平。有一些商业第三方产品支持 ATI 的东西,据我所知,这可能是有史以来最令人惊奇的工具,但它会花费你的钱。
I have no horse in this race; but as someone trying to do my own GPGPU development and who knows a number of other people doing the same, let me observe:
Getting close to peak performance on
the ATI cards is harder, because not
only do you have to break the
computation units up into the very
many compute cores, you actually
have to 4-way vectorize the
calculation on each of the many
cores. For something simple like
encryption or MD5, this may be
straightforward, but it's another
step you have to go through.
The development tools (a debugger and a profiler, plus an entire online development community) that come for free with the NVidia SDK are way more useful than anything that comes with the ATI stuff. It's all well and good to say that OpenCL is OpenCL, but that won't help you find a sneaky bug or find out why your occupancy is lower than you think it should be. There are commercial third-party products out there that support the ATI stuff that may just be the most amazing tools ever made for all I know, but it will cost you.
我在 NVIDIA 的 GPU 上进行 AES,并在 470 上实现了接近 100% 的占用率。我已经使用 OpenCL 有一段时间了,它绝对不是你可以实现性能的东西。它唯一的好处是它的广泛接受度和 x86 支持,但如果您谈论的是性能,请选择 NVIDIA。
注意:如果您可以外包此内容,请给我发送电子邮件:[电子邮件受保护]< /a>
I am doing AES on NVIDIA's GPUs and managed to achieve near 100% occupancy on 470. I have played with OpenCL for a while and it definitely is not something on which you can achieve performance. It's only benefit is it's wide acceptance and x86 support but if you're talking about getting performance, go for NVIDIA.
Note: If this is something you can outsource then send me an email: [email protected]
如果问题不是关于 MD5 哈希的 CUDA 与 openCL(因为这不是关于卡,而是关于你用来实现问题的 api),那么从我在加密货币中看到的情况来看,ATI 硬件更适合哈希算法(openCL 如果我没记错的
话,md5 哈希性能与 SHA-256 相同,因此您可以在此处查看每张卡的哈希性能列表
(哈希/秒)
:哈希 GPU 硬件比较
注意,这是取自 litecoin,它是基于 scrypt 的哈希,但每张卡的最大哈希能力大致相同,只是那个哈希算法改变了计算的难度。
If the questions is not about CUDA vs openCL for MD5 hashing (since this would not be about cards but the api you use to implement the problem) then from what I ve seen though in cryptocurrency, ATI hardware is more suitable for hashing algorithms (openCL implementations)
If I am not mistaken, md5 hashing performance is the same as SHA-256 so you could just check the list of Hashing performance
(Hashes/sec)
of each card here:Hashing gpu hardware comparison
Note that this is taken from litecoin, which is scrypt based hashing, but the max hashing power of each card is approximately the same, its just that hash algorithm that changes the difficulty of calculation.