AMD APU Processors

AMD Accelerated Processing Unit (APU), formerly known as Fusion, is the marketing term for a series of 64-bit hybrid microprocessors from Advanced Micro Devices (AMD) designed to operate as a central processing unit (CPU) and graphics processing unit (GPU) on the same chip.

Development history

The development of "Fusion" technology became possible after AMD's purchase of ATI, a Canadian company, a well-known video processor manufacturer, which took place on October 25, 2006. The technology was originally expected to debut in the second half of 2009, as a successor to the latest processor architecture

In June 2006, AMD employee Henri Richard gave an interview to the DigiTimes website in which he hinted at the future development of the new processor:

Q: What's your perspective on developing a new processor architecture for the next three to four years?

Answer: As Dirk Meyer commented at our analyst meeting, we're not stopping. We were talking about an upgrade of the current K8 architecture, which will take place in 2007. We plan the following improvements to the new architecture: performance of integer operations, performance of real-number operations, memory bandwidth, connections, and so on. You know that our platform is still going strong, but of course we will not stop, and we already have a next-generation kernel that we are working on. I can't give you more details right now, but I think what's important is that we've clearly established that this is a two-horse race. And, as it happens in horse races, even if one horse overtakes the other horse a little bit, that makes all the difference. But the important thing is that it is a race.

In an interview with CRN.com's Mario Rivas, he states: "With the Fusion program, AMD hopes to provide multi-core products using different types of processor units. For example, the GPU will stand out in many parallel computing tasks, while the CPU will take care of the hard work of grinding the numbers. Fusion processors with CPUs and GPUs integrated into one architecture should make the lives of system programmers and application developers much easier."

In April 2009, news broke that AMD had built a trial version of the "Llano" model and was pleased with the results. AMD subsequently postponed the release of the Fusion processor until 2011. Previously, it was believed that the Californian developer will present the processor with an integrated core based on 45-nanometer process technology in early 2010, but the new AMD roadmap delayed the appearance of Fusion for a year, prior to the development of 32-nanometer processor.

At that time AMD was planning two modifications of Fusion - Llano, with four cores and 4 MB of cache memory, and Ontario, with two cores and 1 MB of cache memory. "Llano, built on the AMD Fusion architecture, will consist of four Phenom II-class cores with 4 MB of L3 cache memory and a 1600 MHz DDR3 controller, as well as a Direct3D 11-capable graphics core and a PCI Express 2.0 bus for an external graphics card; these microprocessors will be manufactured using 32-nm processor.

What is AMD APUs?

Architecture Features

AMD APUs have a unique architecture: they have AMD CPU modules, cache memory, and a discrete class GPU, all on the same chip using the same bus. This architecture allows the use of graphics gas pedals such as OpenCL with an integrated graphics processor. The goal is to create a "fully integrated" APU, which AMD believes will eventually contain "heterogeneous cores" capable of automatically handling both CPU and GPU operations, depending on workload requirements.

GPU integration yields a significant increase in bandwidth for the graphics subsystem, reducing power consumption and final product cost. Unlike discrete graphics cards, integrated GPUs have no memory of their own and have to use shared memory.

The advantages of APUs over the classic model of integrating GPUs into motherboard motherboard logic in AMD's vision:

1. Bandwidth between GPU and memory is tripled;
2. No need for some switching nodes;
3. Significant reduction in required wiring;
4. Reduced GPU size;
5. Reduced latency and power consumption.

Expanded interface between the CPU and GPU opens up new possibilities:

1. Advanced voltage management for CPU and GPU;
2. GPU support for higher-level programming languages;
3. Common address space for CPU and GPU;
4. Page memory usage for the GPU;
5. New capabilities for general purpose computing (GPGPU).