找回密码
 新注册用户
搜索
查看: 11274|回复: 20

[新闻] FAH Support for ATI GPUs

[复制链接]
发表于 2011-3-16 10:25:11 | 显示全部楼层 |阅读模式
本帖最后由 金鹏 于 2011-3-16 16:30 编辑

We have been working to improve our support for AMD/ATI GPUs. AMD/ATI originally supported the Brook language for GPU programming, but has since moved to OpenCL, dropping support for Brook. This means that in order for FAH to support ATI GPUs, we have had to rewrite our ATI GPU core in OpenCL. This has been a huge undertaking, but is coming to fruition (check out http://simtk.org/home/openmm for details and a change log with progress updates). Due to ATI's decision to drop Brook in lieu of OpenCL, new FAH ATI GPU core development can only be done in OpenCL and thus can only run on GPUs which ATI supports with OpenCL.

We want to give ATI donors a heads up that ATI does not support all of its GPUs with OpenCL (the series 3xxx are not supported and the series 4xxx does not have sufficient OpenCL support for efficient FAH calculations). For the short term, we will support both the older (Brook coded) ATI core 11, and the newer (OpenCL coded) ATI core 16, but we cannot support the Brook-based core for much longer. We plan to support ATI core 11 until September 1, 2011 (hopefully longer, but we want to get this on donors' radars). Closing support for a given platform is never a popular decision, but in this case, we have no choice due to the lack of vendor support of that platform.

The good news is that ATI's support for OpenCL is very strong and we're excited about our new ATI openCL core (core 16), which is undergoing testing right now. ATI core 16 will be publicly released after the v7 client is released, as the core needs the new client.

评分

参与人数 1基本分 +7 收起 理由
金鹏 + 7 我很赞同

查看全部评分

回复

使用道具 举报

发表于 2011-3-16 10:37:52 | 显示全部楼层
good news
回复

使用道具 举报

发表于 2011-3-16 10:52:30 | 显示全部楼层
大意:
当前A卡的gpu内核(core11)是用Brook写的,它只支持老显卡(2xxx-4xxx),不过由于AMD不再支持Brook语言了,所以我们初步打算在2011-9-1停止使用老内核。
新内核(core16)基于OpenCL,支持5xxx(包含)以上级显卡,现在正在内测,等v7客户端发布后,也会发布。

vmzy 发表于 2011-3-16 10:34
回复

使用道具 举报

发表于 2011-3-16 11:44:00 | 显示全部楼层
斯坦福表示hd5000和hd6000会很猛
回复

使用道具 举报

发表于 2011-3-16 15:54:01 | 显示全部楼层
难道悲剧的4xxx一下就不提供支持了。。。。。。。。
回复

使用道具 举报

发表于 2011-3-16 16:29:28 | 显示全部楼层
ATI's support for OpenCL is very strong

关键是这点是亮点,难道A卡5XXX+开始翻身了
回复

使用道具 举报

发表于 2011-3-16 17:56:44 | 显示全部楼层
终于可以去买点性价比高的卡跑fah了
希望v7能赶快公测
回复

使用道具 举报

发表于 2011-3-16 20:50:37 | 显示全部楼层
按说Brook/Brook+语言的效率应该是很高的 不过这取决于程序员的水平...........
回复

使用道具 举报

发表于 2011-3-16 22:14:38 | 显示全部楼层
按说Brook/Brook+语言的效率应该是很高的 不过这取决于程序员的水平........... ...
JuniperXT 发表于 2011-3-16 20:50

执行一样的指令会快一点,但是如果直接用brook写好程序必须对框架非常了解才能发挥出完全实力
利用opencl写的话,就是把任务交给驱动程序,而驱动是amd自己写的,当然对自己的处理器更了解
回复

使用道具 举报

发表于 2011-3-16 22:29:11 | 显示全部楼层
5000~哦耶!终于翻身了!
回复

使用道具 举报

发表于 2011-3-17 09:00:43 | 显示全部楼层
说是3月份放出来,好期待呀。。。
回复

使用道具 举报

发表于 2011-3-17 11:09:17 | 显示全部楼层
感觉斯坦福的习惯性延迟又要发作
回复

使用道具 举报

发表于 2011-3-17 14:05:38 | 显示全部楼层
不晓得我的4870会怎么样?fah效率太低一直在跑cc和mw。
回复

使用道具 举报

发表于 2011-3-17 14:59:51 | 显示全部楼层
回复 13# eofliyws

2xxx-4xxx系列的应该不会有大的改变 最少是5xxx系列的显卡
回复

使用道具 举报

发表于 2011-3-17 21:30:36 | 显示全部楼层
8楼和9楼有两个高人出现……咋对这种深层次的东西都这么了解呢?
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 新注册用户

本版积分规则

论坛官方淘宝店开业啦~
欢迎大家多多支持基金会~

Archiver|手机版|小黑屋|中国分布式计算总站 ( 沪ICP备05042587号 )

GMT+8, 2024-4-27 17:42

Powered by Discuz! X3.5

© 2001-2024 Discuz! Team.

快速回复 返回顶部 返回列表