There's practically no difference between TDP figures. They are all carefully managed. They set a particular TDP limit, and the hardware will play along that level. It may exceed it for a short-term(responsiveness by Turbo), but long-term duration averages it out so it does not exceed TDP levels.
Also Avoton is ahead of every ARM competitor simply based on that you can go buy it right now. AMD's Seattle won't be available a full year after(Q4 2014/Q1 2015).
The biggest reason Intel isn't playing the mobile market is because they were simply not participating on it. The tables are turned in server.
Because geekbench is just like SPEC CPU 2006 amirite? AMD typically uses gcc and PCG for its SPEC submissions, choosing whichever one gives the higher results. It's very likely that AMD's unspecified compiler configuration gave the Opteron A1100 every advantage possible.
The Cortex-A57 simply doesn't live up to the hype. Big surprise.
There's been a bit of a discussion about this at the real world technologies forum and it looks like there is a particular test libquantum that icc is able to vectorize well but gcc doesn't, so the icc version goes seven times faster. Other than that the results are only a percent different between gcc and icc. Overall the results mean the icc figure is 1.18 times as high. Seemingly the POWER and SPARC own brand compilers also do this vectorization so their figures are comparable.
Even derating the 106 by the 1.18 factor though you get 90 which is still above the 80 that is projected for AMD's Seattle and that at 25W. The main things it seems to have going for it are the two 10Gb Ethernet ports, larger memory capability and the encryption and compression hardware - it is a good mix for a web server.
Market Data 
Markets 
AI
