AMD hopes to take advantage of its technological advantage, and presents architecture Which emphasizes heterogeneous computing
After launching the The Llano processors Officially shed some light on The Trinity processors, Is free AMD Reveal its future graphical core - which promises to be a true revolution in the field of heterogeneous processing (integrating CPU and GPU processing forces to perform computations optimally) - the area that has become the most important for the company.
Starting from the base: VLIW out, SIMD in
The highlight of the company's future graphical core, known as Graphics Core Next or GCN for now, is the neglect of the VLIW (Very Long Instruction Word) design, which AMD relied on since R600 core And the C-Radeon HD 2xxx, for the benefit of SIMD (Single Instruction, Multiple Data) design.
Like VILW, the SIMD method also allows high-quality parallel processing of the graphics cards, but does so with dynamic scheduling capabilities that are missing from VILW (and some other key benefits) - making it better and more suitable for operations Compute generality using the graphical kernel.
The differences between the methods can be described and analyzed across many pages, but the bottom line it presents AMD Says that switching to SIMD will allow Performence Significantly improved for the future graphics core at the time of general calculations, without sacrificing the performance of graphic calculations, which of course remain the central and most important feature for these cores.
|CU - The basic unit by it||We will define the graphical core of AMD soon|
The basic building units to be composed of the GCN cores will be vector SIMD units of width 16 units, with memory of 64KB. Four such units will be added to the 1KB L16 level memory, scalar processing unit, control and decoding units and more, which together form a structure resembling a miniature processor and machine called Compute Unit or CU - the basic computing unit in the future graphical core, which it defines AMD As a combination of MIMD (Multiple Instruction, Multiple Data) capabilities, SIMD and SMT (Symmetrical Multi Threading) capabilities at the same time.
The dynamic scheduling capabilities of the new architecture and design rely on hardware rather than software (which is the case with the VLIW method), which in turn will allow for easier and more intuitive programming for developers, who do not have to worry about efficient scheduling of information, and open the door to use the familiar and popular programming languages Most like C, C ++ and others - where the new graphical cores will be fully supported, which will also allow programming of the processing cores and graphical cores together in a simple and accessible way.
The graphics core becomes smarter and more advanced
The GCN core will also offer virtualization support memory, Using thex8664, similar to general processors, that will allow the sharing of this volume by the various components that make up the graphics core.
Another new feature that will be added to these cores is the ECC Error Correcting Code (ERC) capability, which will join the existing ERC (ERR) Error Detection and Correction capabilities AMD, And will help ensure a minimum of errors and errors during the core graphic work.
|GCN Memory Work -||Almost like a CPU|
AMD has revealed very little detail about the GCN's graphical capabilities, but noted that the core will incorporate Primitive Pipelines that can create two-dimensional and basic 3D shapes that can be increased in size to scale - which should allow Performence Are significantly better during geometric calculations, calculation calculations and other advanced calculations.
AMD also mentioned the ability (called Partly Resident Textures (PRT)) to create and work with ultra-large textures - "mega-textures" that are too large for the temporary memory of the graphical core (Buffer) to be stored in virtual memory, an option that is efficient and significantly faster than hard drive storage Of the system, as has been the case so far.
Why is all that good at all?
The GCN architecture will become the standard for the graphics core AMD - Both on the dedicated graphics cards and on the core of the APU processors, with the goal being clear - the desire to utilize the company's advanced technology not only in graphical processing, but also in general processing, which can certainly be a real advantage in the never-ending battle with Intel.
|A variety of new capabilities for programmers,||And for consumers, hopefully|
Now, with the discovery of its new graphics core, AMD is actually joining NVIDIA, which has launched the The Fermi architecture, Which aims to facilitate and improve the overall processing performance of its graphics core, already before A year and a half.
AMD's current step is a logical and natural continuation of its declarations and attempts to make heterogeneous processing more popular and popular, Fusion processors for the first time.
In fact, even to the third side of this equation, Intel, It is interesting to see heterogeneous processing becoming mainstream, as its processors also offer built-in HD Graphics cores that can benefit from better utilization of GPGPU capabilities.
Larrabee project Intel's ambitious ambition contained several components and aspects similar to what AMD would offer with the GCN cores, and yet Canceling the project, Some of these components have found their way into the graphical cores built into the company's processors, and therefore the development of heterogeneous computing will benefit not only one side, but all companies in the business.
This is a fairly simplistic analysis of the situation, but it is difficult to ignore the fact that heterogeneous computing and GPGPU have very good opening data, which lead us to estimate that the coming period will bring about significant growth for these technologies.
When will we see the GCN on our computer?
Throughout the introduction of the GCN core, AMD was careful not to state a specific time frame or product to use. Many sources in the network estimate that we will see the new architecture already in the company's next generation video cards, Of the Southern Islands family Which is set to hit the market this year, but to us this estimate looks pretty hasty given that production of these cards started a few months ago, in addition to knowing that the next generation of Fusion processors, the Trinity processors, will be based on the VLIW4 method - which strengthens the The assumption that the GCN cores have quite a bit of time to be ready in the spotlight. Where does the truth lie? Until we receive more updates fromAMD, We can only guess.
|Fusion System Architecture:||The future according to AMD|
And what about the home user? How will the new and different architecture affect the performance that consumers will gain? Anyone expecting an instant revolution in the computer world may be disappointed, but in the long run, this "upgrade" will no doubt enable the creation of software that can take full advantage of the GCN cores, in a way that improves the overall performance.