The problem that AMD has right now is caching and memory performance. Considering that in some cases Intel enjoys almost 3x the memory performance and a comparable AMD CPU has this is an issue. Sure you can argue that AMD is working on more GPGPU oriented tasks (which can leverage of the power of the highly parallel GPU core, but considering the limited number of applications that actually use that technology you are aiming at a very small market.

Right now the major applications that are GPU accelerated are items like Flash, Transcoding, and depending on the application video rendering. The reason that these applications can be easily adapted to the GPU is that much of what they perform is very repetitive. You can execute the code in parallel without needing to run multiple streams of different code. It really does make the work load for the GPU easier. Although AMD traditionally has adapted their GPUs to work with smaller blocks of code (called vectored) they have recently expanded the way their GPUs work to remove this block (nVidia used to have the advantage there). This still does not remove the obstacles faced by the fact that many consumer and professional products would make no use of the GPU for compute functions.

Here is where AMD needs to strike a balance between the CPU and the GPU inside their APUs (wow lots of Us there). To do this they have to improve memory performance across the board if they cannot the CPU cores in their APUs will always be starved for data as even with SSDs you are not going to be able to feed a CPU properly. So while AMD can tout great graphics performance in their APUs (and they would be right to) they still cannot compete even at the very low levels with Intel in terms of CPU performance. It is also important to note that as AMD tried to stuff more cores into the CPU component of their APU they are still driving up power needs which will make them even less attractive to OEMs as they continue to look for the best combination of performance and power. Right now for some OEMs the option for an Intel ULV (Ultra-Low Voltage) CPU with Hyper threading and an discrete GPU (even a low powered one) looks better than simply dropping in an APU.

This is a trend that we might see continue as Intel + AMD or nVidia will still maintain the mid to high-end ultra-book/tablet x86/64 market while AMD cleans up in the entry level/value market. That is unless AMD can figure out how to fix their memory problems, if they can do that… honestly they could begin to compete head to head again. If you look back on when AMD was beating Intel they had memory performance on their side. The IMC in the Athlon64 when running DDR could flat out run anything Intel had. When AMD lost footing was when they could not overcome the latency issues with their ICM and DDR2/DDR3 (which was all about proper caching).

AMD will continue to do well with their APUs and will gain back some market share from Intel in the entry level and value market, but until they can balance CPU and GPU performance (read that as fix their caching and memory problems) they will only be able to nibble away at what Intel has on their plate.

Discuss this in our Forum