Exploring the reasons for AMD to develop APU products AMD has officially released its Llano APU products, and the industry has placed high expectations on it. APU is a brand new product that is different from the CPU for decades, and its launch has a deep historical origin. , then exactly why AMD to develop APU products, APU can bring us what is different from the CPU? Today we carry out the following analysis. Of course, my own analysis may not be entirely correct. It is only the author's personal opinions. Users are welcome to submit criticisms.

First of all, APU Chinese name acceleration processor, which combines the previous CPU core and GPU graphics core, and through the industry standard OpenCL and DirectCompute interface for heterogeneous computing, applications will be able to timely call the APU internal CPU and GPU unit co-computing , bringing a faster and richer computing experience. Obviously, AMD hopes to use the APU to make up for CPU deficiencies in order to compete directly with competitors. This intention of AMD has a very deep reason. One of the important factors is that the CPU instruction set has always been mastered by competitors.

The painful lesson forced AMD to rethink the impact of instruction sets on processors, starting with the actions of two processor vendors in recent years.

We know that the processor's instruction set can be considered as a catalyst for processor performance, by which it can greatly accelerate certain computing performance and efficiency of the processor. All along, Intel took the processor market share to lead the development of instruction set extension technology, and cooperated with major software vendors to improve the efficiency of software execution. AMD can only be led by Intel and the product will always lag behind its competitors in instruction set support. Take the 45-nm Penryn support 47 new SSE4 instruction set as an example. Although AMD expressed support for SSE4A, it only contains a few instructions from Intel SSE4.

In order to break the dilemma that instruction set extension technology has been lagging behind, AMD rushed to announce the introduction of SSE5 instruction set extension technology before Intel in August 2007, and plans to deploy it in the next generation "Bulldozer" core architecture after K10, which is expected in 2009. Introduce the actual product. It is understood that SSE5 plans to add more than 100 instructions in the initial planning period, of which the most noteworthy ones include: 3-Operand Instructions and Fused Multiply Accumulate.

AMD previously stated that one of the missions of the SSE5 instruction set is to enhance high-performance computing applications and give full play to the advantages of multi-core, multimedia parallelism. SSE5 will introduce functions that used to exist only in high-performance special architectures into the x86 platform to maximize the output capacity of each instruction and enhance the code base.

In August 2007, AMD announced the SSE5 instruction set (previously from Intel SSE to SSE4), but Intel immediately stated that it would not support SSE5. In March 2008, Intel announced the Sandy Bridge microarchitecture, which will introduce a new AVX instruction set. In April, Intel announced the AVX instruction set specification and then began to update it.

Intel’s move has greatly influenced the future software vendor’s support orientation. AMD’s influence in the industry is far less than that of Intel. If AMD unilaterally favors SSE5, it is bound to lose in the future performance competition; under pressure from competition, AMD cannot Do not choose to support Intel's proposed AVX (Advanced Vector Extensions) instruction set, while using the AVX framework to rewrite the AMD SSE5 instruction set, redefined as XOP (eXtended Operations Instruction Extension), CVT16 (half-precision floating-point conversion) and FMA4 (4 Operand multiplication plus).

Making this decision is costly, which means that the processor architecture needs to be substantially adjusted. By November 2008, AMD's latest roadmap shows that Bulldozer has been postponed until 2011. AMD's bulldozer processor can be said to have plunged a big mistake. This has prompted AMD to bypass the CPU and use its own advantages to make up for its shortcomings. Obviously this task falls on the GPU.

APU V. S. CPU: open challenge arbitrariness In addition, AMD development APU has a very large application development background, because of our daily computer use, will come into contact with more and more 3D computing, parallel data information, such as our future web pages will be to 3D To develop, our video information needs high-speed parallel processing. This kind of calculation is the advantage of GPU. This demand has also become the basis for the development of the APU.

The CPU itself has a high operating frequency and a small number of cores. If a large-scale parallel computing is performed, it is obviously inferior to a GPU. Although the processor instruction set can be optimized to speed up the computational performance, the future development needs of users are increasing. Changes, and the instruction set is firmly occupied by the Intel family, it is obviously impossible to achieve optimization through fully open programming based on application development.

At present, GPU technology has been developed to DirectX11, and there are already open industry interface standards such as OpenCL and DirectCompute. If the software supports OpenCL and DirectCompute in the future, the acceleration effect it brings may be better than that of the processor instruction set. The importance of the command set for a solitary car may not be as high as it was in previous years.

It needs to be mentioned: Intel's Sandy Bridge processor only supports DirectX 10.1 and cannot support the OpenCL and DirectCompute recognized standards.

APU OpenCL Accelerated Application Accelerated Application AMD announced on June 13, 2011 that more and more innovative software applications are adopting the OpenCL industry standard on a large scale. AMD has worked closely with leading software companies to help them optimize applications on common platforms while also accelerating these programs with the latest technologies including the AMD Fusion Acceleration Processor (APU).

Software developers are happy to adopt common application programming interfaces (APIs). Thanks to the flexibility of the cross-platform, cross-operating system, and cross-hardware vendors inherent in the OpenCL standard, the software industry is embracing a wave of consumer and business applications based on the OpenCL standard.

The following are examples of leading applications that already support OpenCL or support OpenCL in the near future:

DVD/Media Players • ArcSoft: TotalMedia Theatre – All-in-One Video Player • Corel: WinDVD—Blu-ray and DVD Player Software for Telepresence and Webcam Applications • ArcSoft Company: Webcam Companion—Combination of 3D and HD Webcams Applications • ViVu: VuRoom - Multi-party desktop videoconferencing software • ViVu: VuCast - Wide-area video webcast software Video production/editing software • ArcSoft: ShowBiz - video editing software • Corel: Digital Studio - integrated multimedia software Group • Corel: VideoStudio Pro - HD Video Editing Software • Cyberlink: PowerDirector - Video Editing Software • Sony: Vegas Movie Studio HD - Home Theater HD Video Editing Solution • Sony: Vegas Pro - Professional HD Video, Audio And Blu-ray Disc Production Solutions for Video, Photo Effects, Imaging, and Utility • ArcSoft: Panorama Maker Pro - Photo and Video Stitch and Media Managers • Eyeon Inc.: Fusion - Visual Effects and Synthesis Tools • HP Labs: " One Page is Different (Every Page Is Different) Raster Image Processing and Gigapixel Real Time Imaging for HP's Large Format Business and Industrial Printing Solutions • Viewdle: Uploader - Photo Facial Recognition Software • Viewdle: Video SDK - Video Development Face Recognition Software Toolkit Video Conversion Software • ArcSoft Corporation: MediaConverter - Multimedia File Converter • Rovi: MainConcept Transcoding Platform - Professional Transcoding Application • Rovi: MainConcept H. 264/AVC OpenCL - Coded Software Development Kit Engineering Simulation Software • Altair Engineering Company: HyperWorks RADIOSS - Linear and Non-Linear Simulation Finite Element Analysis (FEA) Solvers • Dassault Systemes: Product Lifecycle Management (PLM) - Simulation and CAD Software • DEM Solutions: EDEM - Discrete Element Modeling Software for Particle Flow Simulation • ESI Group: PAM-CRASH and PAM-STAMP 2G Solvers - Prototyping and Manufacturing Process Digital Simulation Software • MSC Software: MSC Nastran - Universal Finite Element Analysis Solutions • OPTIS: RTLab and VRLab - Real-time ray tracing software solution AVX2 has debuted Intel needs to strengthen GPU

Recent foreign news shows: In 2013, Intel's next-generation architecture processor Haswell will support the new instruction set "AVX2", AVX2 will add more functions based on the AVX instruction set, thereby improving the processor's computing performance.

Intel still improves its processor performance by updating the AVX instruction set. However, obviously, Intel also needs to improve its graphics performance. For AMD, developing its own heterogeneous computing architecture will be more appropriate than optimizing the instruction set in the precarious state. , otherwise it will be held up by competitors for a long time and it is in a passive situation.

Haswell will not appear until 2013. Maybe when AMD's heterogeneous computing has blossomed, the importance of AVX2 may not be so important.

The arrival of the APU means that the new round of competition between the CPU and APU, Intel and AMD is starting. Hope that APU will go all the way to bring better products to consumers.