来自HARDOCP的DX10文章（围绕ATI幻灯片）

Edison

http://enthusiast.hardocp.com/ar ... CxoZW50aHVzaWFzdA==

Base DirectX 10 on a Unified Architecture


DirectX 10 is built from the ground up with a unified architecture in mind. This unified API and GPU architecture collapses the traditional vertex, geometry and pixel units into one processing stage instead of separate units. So, the API can handle all of these instructions through a single GPU stage instead of it having to be broken apart as it has been in the past. Instead of vertex, pixel, and geometry shader units, just think one big unit to do it all. Today we are at many times seeing processing units on a GPU being wasted due to applications bottlenecking other units. Currently we might see Vertex Shading units going unused while waiting for Shader Units to complete their processing (or vice versa). Unified architecture should remedy much of this latency and wasted silicon cycles.


While DirectX 10 can fully utilize a unified GPU architecture, a unified GPU isn’t required to meet DirectX 10 specification. Therefore you can still have a GPU with separate vertex, geometry, and pixel units that is completely DirectX 10 capable and compliant. The key component that will need be to supported by a non-unified GPU is the new Geometry Shader that could work in a fixed pipeline architecture as shown above.


One of the key ideas behind a unified architecture is to move the GPU from a rendering only processor to a complete compute processor. Right now all the GPU does is render 3D and displays it on your screen (yes it does more like 2D, video etc... but for the point of this article we are talking about 3D). With a unified architecture the GPU becomes more. It becomes a processor that can do almost anything that needs code processed. This means the GPU can take on more functions like physics, AI, animation and many other processes that can benefit the gaming experience. DirectX 10 and a unified GPU architecture helps a video card become an all-in-one Swiss army knife of game processing. Those are the ideas at least, how it all works out is up to the game content developers.


In the slide above you can see there is a new thread arbiter. This arbiter unit issues threads to the core which determines how much of the scene needs vertex, geometry or pixel shading resources. It basically plays “traffic cop” for all of the instructions moving towards the GPU. The GPU then does all the required work rendering your scene. This will allow highly complex vertex and pixel shaded-heavy scenes to be processed with the full resources of the GPU. The thread arbiter can issue many threads at once, just like multi-threading with a CPU. You can see how all of this ties in with the “Ultrathreading” core introduced in the Radeon X1000 series technology. The X1000 series could be considered a hybrid Unified Architecture, as it is a step right before a fully realized Unified Architecture.

You can now see how all this ties into the XBOX 360 as well. The XBOX 360 uses ATI GPU technology and it is a unified architecture. Therefore ATI already has a 1st generation unified architecture GPU on the market. Obviously the XBOX 360 is ahead of desktop GPUs in terms of this technology.


ATI’s next generation desktop GPU will be a fully unified architecture. Since the XBOX 360 is ATI’s first generation unified architecture technology their next generation desktop GPU will be ATI’s 2nd generation unified architecture technology. That means ATI has taken what they learned from the XBOX 360 unified architecture and made improvements to create an even better unified architecture for the PC with their next generation desktop GPU. This also means porting games between the XBOX 360 to ATI desktop GPU hardware should be easier.

51HDTV

好长,有全文翻译过来的吗?

letitbe

This also means porting games between the XBOX 360 to ATI desktop GPU hardware should be easier.

对这句话很感兴趣，这将意味着以后XBox游戏移植到PC会更加容易（反之亦然）

fspgp

看来未来的趋势真是多核cpu代替图形加速卡。这样还省事了，不用操心显卡了　！＃