Menu Content/Inhalt
AMD APU in 2013 Print
April 2013

APU is a term coined by AMD to stand for Accelerated Processing Unit.  APU consists of a CPU and a GPU.  This composition was not different to Intel Sandy Bridge CPU released nearly a year earlier.  Though it is not the first to the market, APU has its technical merits.  This article briefly explains how APU was positioned and how it attempted to differentiate from Intel in the latest development in 2013.

Background

o AMD has kept the CPU Only approach under various brands including FX, Phenom, Athlon, and Sempron (in descending performance) in parallel with APU.  These CPU models do not have GPU inside.   They share the same socket form factor and pin counts, called AM3 or AM3+ depending on the TDP (thermal design power) limit.  Phenom has ceased in 2013.  

o AMD released A4, A6, and A8 series of APU in 2011 and A10 series in 2012.    They have a separate socket to AM3/AM3+ and it is FM1 in 2011 and FM2 in 2012.  The following models are the top of the line in the year listed.

2011 FM1 A8-3850 4C 2.9Ghz 4MB CPU, Radeon 6550D 400C 600Mhz 
2012 FM2 A10-5800K 4C 3.8Ghz 4MB CPU, Radeon 7660D 384C 800Mhz
2013 FM2 A10-6800K 4C 4.1Ghz 4MB CPU, Radeon 8670D TBA 844Mhz

Technical Performance

o AMD positioned the above top of the line APU models against Intel Core i3 in terms of computational performance.  Nevertheless various 3rd party benchmarks showed APU to perform better than i3 as soon or as far as graphics or DX11 is involved.

o AMD claimed to do better in screen counts: R-series APU has provision for connecting to 4 screens.  If the same PC has a discrete Radeon card capable of supporting 6 screens, the APU screen provision will stay whereas it is not the case for Intel implementation.  An APU-based PC is capable of supporting 10 screens.  (Note: no info on R-series)

Real Colour in GPGPU

o AMD made an announcement in March 2013 about using the GPU integrated in the same die of the CPU for general purpose computing rather than for display.  The announcement mentioned the Teranium Project which is a 4x 10Gbit (for 4 ports) network interface driver for a deep packet inspection extensible framework.  This framework bypasses TCP/IP software layers of the Linux core and handles TCP/IP within the integrated GPU.   This approach produces higher performance than a discrete and powerful GPU card in the computer by a big margin.   The reason is simple.  The discrete approach relies on the PCIe bandwidth for CPU-GPU data transfer and coordination.  The integrated approach relies on the internal CPU die bus which has a higher bandwidth than PCIe.

o To implement this approach, AMD applied OpenCL 1.2 programming standard (and OpenGL 4.x for graphical display).  It was the first public announcement by AMD on applications using the integrated GPU for non-display purposes. 

END