蛮可怜的

喷了 原来是读写加起来192GB 那实际gpu频率=96GB/128bit=750MHz 比原先800少了50MHz ms自黑？

引用:

原帖由 你老闆 于 2013-6-29 02:43 发表

To the best of their knowledge, 800MHz remains the clock speed of the graphics component of the processor, and the main CPU is operating at the target 1.6GHz. In both respects, this represents parit ...

To the best of their knowledge
ms当然不会告诉他们降频了 我不清楚内存带宽是否能把读写加起来算 就算可以 原来是102 那也该是102X2=204 怎么才192？
老外有个解释
Right, that 204 is the theoretical max if they could just read and write wherever they wanted on every clock cycle. A real-world 133 GB/s would be much more realistic.
However, they gave 192 GB/s as a theoretical max. If this is the accurate figure, there are precisely two possibilities why:
1-The methodology they use to simultaneously read and write to the eSRAM is mathematically proven to work, at best, on ~8/9 clock cycles. This would explain the 1.88 factor of increase, as even in a perfect world you could not double the bandwidth.
2-The theoretical max assumes with perfect optimization read/write is always possible, but the GPU has been downclocked 50 MHz so the bandwidth in each direction is now 96 GB/s.
The first scenario is less likely, imo, because 8/9 is a really weird ratio for a binary system. Also, there is very rarely a reason to do anything but double the max rate of theoretical access if you discover sometimes simultaneous read/write is possible. In theory, one should be able to read and write completely separate memory adresses exclusively, which would truly double the rate.
I am not an expert, and the article may not be accurate, but this sure looks a lot like a downclock.

引用:

原帖由 KoeiSangokushi 于 2013-6-29 11:14 发表
posted by wap, platform: Chrome

两种可能，一种是降频到750MHz，读写都达到96GB/S
另一种可能，不降频，192GB/S是说读写不能都达到102GB/S，比如读达到102GB/S了，写就只能90GB/S了

本帖最后由 KoeiSangokushi ...

没可能像你说的这样工作 只能除以2 这是二进制系统
另外 按照我的理解 内存应该是独占的 怎么可能读写同时进行 这会引起内存泄露

引用:

原帖由 1320581 于 2013-6-29 11:44 发表
索饭巴不得频率降到0，这样ps4好一统宇宙:D

软饭恨不得数学公式出错？喷了

喷了 还在这里扯gddr5的延迟不适合做内存……这挣扎的太难看了

引用:

原帖由 KoeiSangokushi 于 2013-6-29 14:13 发表
posted by wap, platform: Chrome

192/102.4=1.875
也就是1又7/8
也就是可能当完全读取102.4GB/s带宽的时候，有4MB的ESRAM不能在此时被写入（28/32=7/8），所以此时写入带宽仅为89.6GB/s
完全符合，不需要降频

我觉得你还是相信降频的好 哪有这样设计的硬件？xbone sdk是不是还要特意注明一句 使用esram的时候 请判断写入的是28还是4

引用:

原帖由 讴歌123 于 2013-6-29 19:27 发表
posted by wap, platform: HTC (8X)

TitanFall有没有单机是不是全程联网搞清楚没？普及云计算的文章写起来没？话放出来的要说到做到啊……

有单机？我没兴趣看新闻 你告诉我啊
我说有空才写 而且就你一个人对所谓云计算这么感兴趣 我实在没动力写
88% more为什么不是100% more 期待大师解答

引用:

原帖由 KoeiSangokushi 于 2013-6-29 20:22 发表
posted by wap, platform: Chrome

192/102.4=1.875
也就是1又7/8
也就是可能当完全读取102.4GB/s带宽的时候，有4MB的ESRAM不能在此时被写入（28/32=7/8），所以此时写入带宽仅为89.6GB/s
又或者是ESRAM的读取可以 ...

xbone开发者会疯 能控制到时钟周期的代码得是什么样 我无法想象 汇编似乎都达不到