PCEVA,PC绝对领域,探寻真正的电脑知识
打印 上一主题 下一主题
开启左侧

讨论帖:RX 480的4CUs到底拿来干什么?

[复制链接]
1#
chungexcy 发表于 2016-8-15 23:14 | 显示全部楼层
AMD的Polaris白皮书其中有一页写到RX 480有预留4个CU单元用作某些音频处理任务,而下边的注释又加了一句当这4个预留的CU单元在使用时,是无法做其他任务的,
按照白皮书上的那段描述,可以确定的是一共只有36cu,其中4个cu能作为音频渲染。不过单从文字上看,确实无法断定这块4cu是不是被划死了,估计只有开发文档可能说的更具体一点。(我的理解是动态分配的)。不过无论怎样,这块的控制权限肯定是开发者的,开发者可能会在游戏中给这么一个开关,但不会直接给用户一个全局控制。

real-time audio raytracing真要做好,我个人理解计算量不低。amd想借这一招棋走好vr,至于是否是一招好棋只有让时间来说话了。不过不要忘了ps4 neo和psvr的存在,这两对主导游戏发展的走向起到很大的影响。

顺便再引用一下,AMD对Trueaudio的官方说明
How AMD TrueAudio removes constraints on sound developers

AMD TrueAudio technology is all about giving sound engineers the freedom to follow their imaginations and the power to make their games sound as convincing as they look. Here are some of the features that make that possible:

  •     A dedicated digital signal processor (DSP) is built in to the AMD GPU core. That’s a hefty dose of processing power committed just to generating immersive soundscapes. That doesn’t just enable new and exciting audio features in games; it also saves CPU cycles that can be used for other tasks.
  •     Programmable sound effects bring to audio the same kind of flexibility that programmable shaders brought to graphics. Game developers aren’t stuck with inflexible canned effects anymore. They now have the flexibility to create complex effects and acoustic environments.
  •     More voice channels and audio objects mean game developers no longer have the unenviable task of determining which sounds are expendable. AMD TrueAudio technology multiplies the number of sounds a game can generate at once, giving developers the capacity to create much more lifelike soundscapes.
  •     True to life echoes and convolution reverb are finally available on all platforms, meaning programmers can now build these effects into their games rather than relying on basic reverbs. The types of real-world acoustic phenomena that can be faithfully reproduced are massively greater.
  •     Multi-channel spacialization brings in-headset surround sound with accurate positional audio algorithms to all gamers; not just the ones with the priciest headgear.

5个里面,后面3个新特性明显是给vr用的:多声源融合,回声反射(混响)模型,精确多音源定位。个人理解是amd把这些都做好了,留出接口给开发者调用。按照wiki上的说法,GPGPU编程确实会方便许多。
A new version of TrueAudio, TrueAudio Next, was released with the AMD Radeon 400 series GPUs. TrueAudio Next utilizes the GPU to simulate audio physics. The move from a dedicated DSP to GPGPU breaks compatibility with the previous TrueAudio implementation.
个人认为,一个新功能的性能其实并不是最主要的因素,好不好用,通不通用才是影响开发者选择的关键。


2#
chungexcy 发表于 2016-8-16 00:10 | 显示全部楼层
奇趣怪谈 发表于 2016-8-15 23:31
是的,感觉AMD在下一盘大棋。。。
btw如果是需要在游戏控制,那肯定需要全新的API直接调用那4个CU单元了.. ...

应该不是直接调用cu。直接调用的是API,比如告诉声源相对的3d位置和音频内容,然后得到左右声道的波形。API内部由显卡完成调度和计算,这个对API使用者是黑盒子一般不用考虑。
如果这个真能成为规范,到时候应该会有引擎的,让开发更脱离底层。
3#
chungexcy 发表于 2016-8-16 09:50 | 显示全部楼层
本帖最后由 chungexcy 于 2016-8-16 09:51 编辑
rushmytowers 发表于 2016-8-16 09:39
究竟a猫d有没有40cu的poalris??

没有,具体看楼主引用的架构图。。。写新闻的人,数学是体育老师教的,4x9=40。。。
您需要登录后才可以回帖 登录 | 注册

本版积分规则

快速回复 返回顶部