AMD Accelerated Processing Unit
- AMD
- GlobalFoundries
- TSMC
- Socket FM1
- Socket FS1
- Base FS1r2
- Base FP2
- Socket BGA -413
- Socket FM2
- Socket FM2
- AM1
- Bobcat
- Husky ( Llano )
- Piledriver ( Trinity, Richland )
- Jaguar
- Steamroller ( Kaveri )
AMD Fusion is united on one die of code and brand name of a processor concept, the CPU and GPU, and video and other hardware accelerators. It is a result of the acquisition by AMD ATi. AMD calls this construction Accelerated Processing Unit ( APU). The first models based on this concept for use in netbooks and similar devices were presented in January 2011, followed by others in the course of 2011. They are part of the HSA program, the HSA Foundation.
- 5.1 Llano (K10 architecture, Husky core)
- 5.2 Trinity ( Bulldozer architecture, Piledriver core)
- 5.3 Richland ( Bulldozer architecture, Piledriver core)
- 5.4 Kaveri ( Steamroller architecture, Steamroller core)
- 6.1 Bobcat - based models ( Zacate, Ontario ) 6.1.1 Zacate E - series
- 6.1.2 Ontario, C Series
- 6.1.3 Ontario, Embedded G-Series
- 6.2.1 Kabini, A- and E-Series
- 6.2.2 Temash, A- series
- 7.2.1 Llano, A- and E-Series
- 7.2.2 Models for the desktop
- 7.2.3 Models for notebooks
- 7.3.1 Trinity and Richland, A- series
- 7.3.2 Models for the desktop
- 7.3.3 Models for notebooks
- 7.4.1 Kaveri, A- series
- 7.4.2 Models for the desktop
History
AMD demonstrated its first Fusion APU on 1 June 2010 at Computex. The demonstration included, inter alia, a short recording that versus Predator showed a section of the 3D game Aliens, which should be run on an Ontario system in real time.
Concrete products in the form of the E and C series were presented on January 4, 2011. The Llano series for notebooks was presented on 14 June 2011.
On May 15, 2012, the second generation of mobile A- series called Trinity was made (based on the Piledriver processor core of the Bulldozer architecture ) public. This replaces the Llano series, which is still based on the old K10 architecture.
Marketing
The processors or APUs have no marketing names as often applied as with Phenom or Athlon. The only brand name in the processor name is "AMD ". For all AMD systems but there is still the AMD Vision marketing name. Depending on performance and functionality appropriate vision stickers are also accessories such as "Smart HD", for the lowest and weaker versions, on " Brilliant HD Everyday" and " Brilliant HD Entertainment ," for the lower mass market, to " Brilliant HD Performance " for the top ," adds mainstream "area.
Technical implementation of the concept
Core aspect of the Fusion technology is the direct connection of essential system components - x86/AMD64-Prozessorkerne, Vector Engines ( SIMD ) and Unified Video Decoder (UVD ) for high definition video playback - on the same high- speed bus with the system main memory ( random access Memory or RAM). The architecture is so some disadvantages bypass with integrated graphics processors ( IGPs ) are connected in previous single-chip solutions, such as increased memory latency and power consumption, and lower running time on battery power. AMD calls this construction Accelerated Processing Unit. The multi-core processors should contain one or more main processing cores (CPU) and at least one additional processor for specific tasks, for now a graphics processor (GPU). This combination is then connected together.
So far there is the part of AMD following reactions of the fusion approach for different areas:
Sub-notebooks and tablets, netbooks and nettops
Ontario and Zacate ( Bobcat architecture )
Bobcat is the code name for the architecture of a dual-core processor with integrated GPU and Northbridge that has been optimized for low power consumption and low price and therefore has relatively low processing power. Areas of application are favorable systems such as netbooks and nettops as well as devices which are intended to have very low power dissipation, such as sub-notebooks and tablets. When Bobcat is in contrast to the competing product Intel Atom to a more efficient out-of- order processor architecture, which forms the basis for AMD's Ontario and Zacate APUs, which in the series C, E and G (AMD Family 14h Processor) be used.
Kabini and Temash (Jaguar architecture )
The Jaguar architecture replaces the Bobcat based processors. It forms the basis for AMD's Kabini and Temash APUs series A and E. This generation 28 -nanometer Bulkprozess is converted to TSMC, which allows a core area of 3.1 mm2 (for comparison: Bobcat in 40nm process 4.9 mm2), thus reducing power consumption and the area per core. This allows up to four cores, which can be summarized in a so-called module. Due to the modularity of the system, similar to the Bulldozer architecture, one can better respond to customer needs (see Sony's PlayStation 4). The size of the processor cache, which is shared by all cores ( shared), increases to 2 MB, and the floating point unit works with 128- bit data width. Pairing is the Jaguar cores with an "Graphics Core Next " graphic core ( " GCN " ), which means that much more graphics power available than its predecessors.
With Jaguar AMD support for the first time in low- voltage processors the complete SSEx instruction set and also AES and AVX. This command sets were previously only the big architectures such as bulldozers ( SSEx, AES and AVX ) or K10 (only SSE4a ) reserved. The IPC ( instructions per cycle) are expected to rise by about 15 %.
Notebooks and Desktops
Llano (K10 architecture, Husky core)
Llano is the codename for a processor architecture with an integrated GPU and Northbridge, which is designed for the lower "mainstream" area and in notebooks and desktop computers than are eligible. This is classified as AMD AMD Family 12h Processors. The Fusion Llano APU combines two to four Husky cores K10 - generation (AMD Family 10h Processor) with compatibility with x86 instruction sets and the AMD64 architecture, and a DirectX 11 -compatible graphics core with Radeon HD like it 5570 Cards can be found. In contrast to the Radeon HD 5570 UVD has already been integrated UVD 3.0 instead of 2.0 in the graphics chip.
The parallel processing power of the GPU part is intended to supplement next to the graphics acceleration via application programming interfaces such as OpenCL, WebGL, AMD APP (formerly " ATI Stream" SDK ) and Microsoft DirectCompute, the serial processor cores just in Gleitkommabereich.
Although the scalar x86 cores and the SIMD engines of APUs share a common path to system memory is still separated at this first generation of memory in different regions. On the one hand there is the managed by the operating system memory area, which runs on the x86 cores, on the other hand by the software which is running on the SIMD engines, managed memory regions. For the data exchange between the two parts AMD has set up high-speed block transfer engines. Unlike data transfers between external frame buffers and system memory these transfers should never occupy the ( external ) system.
Trinity ( Bulldozer architecture, Piledriver core)
The APU, codenamed Trinity replace the Llano series with K10 inner life. The newer architecture with GPU and northbridge, also designed for the mass market, is also used in notebooks and desktops. It combines the processor technology Piledriver CPUs ( an optimized version of the AMD Bulldozer architecture (AMD Family 15h Processor) ), designed as modules, with newer Radeon HD GPUs. This is around cores similar to the Radeon HD 7350-7670 the AMD Radeon HD 7000 series. Market launch for notebook CPUs was the 15th March 2012, the desktop CPUs were presented on October 2, 2012.
Richland ( Bulldozer architecture, Piledriver core)
The Richland APUs in turn solve their predecessors from the famous Trinity basis. Despite new code name they put on the same architecture and offer just a new stepping. The GPUs are similar to those in the AMD Radeon HD 8000 series ( a new edition of the HD 7000 chips for the OEM market ). The update serves as an interim solution for the 2014 coming out Kaveri APUs that rely on Steamroller architecture.
Kaveri ( Steamroller architecture, Steamroller core)
The Kaveri APUs follow the Trinity and the Richland series. They were announced on January 14, 2014 and included the new Steamroller architecture, which is a far-reaching revision of the Bulldozer architecture. The integrated GPU is on the GCN architecture, as it was presented with the Bonaire chip in the Radeon HD7790 in March 2013, converted. It should come in the course of 2014 APUs for desktop computer ( with socket FM2 ), notebooks, in the embedded and server area on the market.
In moderate resolutions such as 720p games run on a Kaveri system mostly liquid with 40 or more FPS. This requires a faster main memory ( dual-rank DDR3 RAM from 2.400 MHz clock ), since the memory bandwidth is the limiting factor. Another performance boost should arise with the new graphics interface Mantle AMD in the Catalyst graphics driver.
Specifications - nettops, netbooks, subnotebooks, Tablets
Bobcat - based models ( Zacate, Ontario )
The available memory bandwidth (1- channel DDR3 - 1066 or DDR3 -1333 with 64 -bit memory width) is shared by the CPU and GPU in concurrent access. The actual chip area ( the size) is 75-77 mm2.
Zacate E - series
Zacate is AMD's codename. For a 18 -watt APU for the mainstream notebook market in 40 - nm technology The models have the advantage over the C-Series a higher clock for both the processor and the graphics core.
- Memory support: DDR3 SDRAM, DDR3L SDRAM ( single-channel )
- Connection to the chipset: UMI 2.5 GT / s
Ontario, C Series
Ontario is the code name for a dual-core system-on -a- chip implementation in 40 nm technology. The APU integrates the Bobcat core processor and is designed for ultra-thin notebooks, netbooks and other products below the 20 -watt limit. In an Ontario APU plug one or two Bobcat cores and a DirectX 11 graphics core with 280 MHz. The BGA package of optimized for mobile applications Ontario is designed for soldering on motherboards for thin-and -light notebooks and netbooks.
In early 2011, the series was introduced with two models. Has the single-core version of C-30 doing a clock of 1.2 GHz for the processor core, the dual-core version of C-50 1 GHz for both cores. The dual-core versions C -60 and C -70 have according to current knowledge, exact same specifications. According to AMD, the Radeon branding the GPU was only changed on the SoC in order to bring it into line with the other products in addition to the APU name.
- Memory support: DDR3 SDRAM, DDR3L SDRAM ( single-channel, up to 1066 MHz)
- Connection to the chipset: UMI 2.5 GT / s
Ontario, Embedded G-Series
With the Embedded G-Series platform makes AMD Fusion technology in early 2011 available for embedded systems, it is more Ontario versions. The APUs integrate a package on a surface of 361 mm2 ( 19x19 mm) or two 64 -bit processor cores Bobcat class and a DirectX 11- capable graphics unit that can be used as a vector processor. Since March 2011, AMD also offers so-called " headless" variants for embedded systems without video output; these have the same package area of 361 mm2.
Jaguar - based models ( Kabini, Temash )
Kabini, A- and E-Series
- Memory support: DDR3 SDRAM, DDR3L SDRAM ( single-channel )
- Connection to the chipset: UMI 2.5 GT / s
Temash, A- series
- Memory support: DDR3 SDRAM, DDR3L SDRAM ( single-channel )
- Connection to the chipset: UMI 2.5 GT / s
Specifications - Desktop and Notebook
Jaguar - based models ( Kabini )
Desktop Athlon and Sempron models
Husky - based models ( Llano )
Llano, A- and E-Series
The APU Series A and E were published in the summer of 2011. They are primarily intended for the mainstream and low -end systems in the notebook and desktop segment. On a silicon die two to four x86/AMD64-Husky-CPU-Kerne are based on the K10 architecture combines with improved memory controller and a DirectX 11 -capable GPU.
The APU is manufactured in a 32 nm SOI process of Globalfoundries and aim for the same target markets as the Athlon II line. Furthermore it has the combined processor an integrated PCIe 2.0, a dual-channel DDR3 -1600 memory controller, and 1MB L2 cache per core but no L3 cache. When equipped the motherboard with only one memory module per channel memory controller of desktop processors from the A8 and A6 series is specified for DDR3 -1866, in mobile processors with up to 35W TDP is maximum DDR3 -1333 provided. The communication with the chipset / southbridge via the Unified Media Interface ( UMI ), which is based on PCIe, with 5 GT / s ( Giga Transfers / second).
Models for the desktop
Models for notebooks
Piledriver - based models ( Trinity, Richland )
Trinity and Richland, A- series
This second generation of the A - series was 2012 ( mobile processors ), and the desktop models on October 2, 2012 ( desktop) published on 15 May and at Computex 2012 announced ( desktop). It is based on the Bulldozer architecture based with Piledriver cores. The GPU part uses a 4D VLIW shader design, which was introduced with the Radeon HD 6900 graphics cards. Like Intel, AMD now marked processors with an upwardly open multiplier by a "K ", which is appended to the model number. CPUs with disabled graphics unit will be marketed under the old name " Athlon II ", even if the new processors now have a different architecture as a base.
On March 12, 2013 AMD introduced the second generation Piledriver based APUs (ie, the third -generation A - Series) with names Richland ago. Except for minor changes, speed increases and a presumably new Stepping these APUs remain the same compared to the Trinity predecessors.
Models for the desktop
Models for notebooks
Steamroller - based models ( Kaveri )
Kaveri, A- series
The fourth generation of the A - series was announced on 12 November 2013, the AMD Developer Conference APU 13 launch was on 14 January 2014. The beginning makes the desktop area of the A10- 7850K, the two Steamroller CPU modules four threads processed in parallel.
The GPU part has a Radeon graphics unit comparable to those of the AMD Radeon R200 - R7 series models ( " Volcanic Islands "). It is based on the GCN architecture ( "Graphics Core Next " ) version 1.1. This was introduced with the Bonaire chip, the AMD Radeon HD 7000 series. The GPU takes up almost half of the die area to complete and is compatible with DirectX 11.2, OpenGL 4.3 and the AMD - powered 3D interface Mantle.
The APU has a first time in the desktop segment unreacted heterogeneous system architecture ( Heterogeneous System Architecture, or HSA), which is to improve the cooperation between CPU and GPU GPGPU addition. Here, the graphic part will be more intense than previously used to support the processor cores in its calculations. This includes the direct exchange of information without going through an operating system ( Heterogeneous Queuing or hQ ) and accessing the same address space of memory ( Heterogeneous Uniform Memory Access or HUMA ), so that the calculation results are both parts quickly available. In addition, an audio coprocessor is integrated into the APU.
Using Custom Thermal Design Power ( CTDP ) can in the BIOS the maximum power dissipation of the CPU are configured, this allows eg A10 - 7850 to operate with 45W or 65W. The clock frequency of the CPU is throttled depending on the CTDP to specific values . Is actively promoted this at the moment only for the A8 - 7600, which is explicitly listed with 65W and 45W.