To: neolib who wrote (264306 ) 8/11/2010 4:06:10 AM From: pgerassi Respond to of 275872 You want to talk of old CPUs that are still around and are good enough for gadgets and many other things, the venerable Z80 has been made by the billions for decades. It is in toasters, industrial controls, musical instruments and test instrumentation among many others. And its extremely low power now on a 55nm process with speeds over 50MHz. Heck 55% of CPUs made are 8 bit microcontrollers. And since about 10 billion CPUs are made each year, ARM is quite a ways back in the numbers department.en.wikipedia.org Don't count x86 out either as Ontario will really cover where ARM can go. As most designers know its the required power of the back-light on LCD displays that is the major power hog. On larger than 4" displays, it consumes more than half of the power. By the time you get to 10" displays, the back-light makes a 4W power difference of the CPUs (4.5W vs 0.5W) mean less than a 5-10% difference in battery life. To display full HD (1080p) you need at least a 7.1" and typically about 15.4" display. For 1280x720 you can get down to 5.2" and typically 9.2" so for HD video count on a 10" screen minimum. This makes the power draw of a CPU/APU (GPU) combo largely irrelevant. Ontario with 2 1W cores, a 1W GPU core and 1W uncore (1-1.2GHz max vs 400-600MHz idle with the 40SP (80?) APU at 500-600MHz vs 200MHz idle) should be quite favorable from a battery life standpoint for 10" displays wrt performance vs ARM. CoreMark is pushed by ARM because it fits in their small L1 caches, but on Ontario both Bobcat CPU cores and the 40 (80?) SP cores of the APU can run it for a 48,000 score at 1GHz/500MHz far above the 4,000 claimed for the 1GHz ARM A9 Cortex DC. Better than the CoreMark/MHz or CoreMark/W ratings of 1GHz ARM A9/DC too. I submit that this benchmark is irrelevant anyway given that most applications are much bigger than ARM's cache size and need to be benched with that in mind. As for your contention that ARM graphics will advance in a power efficient manner, that is nothing but a hope and a prayer, being far behind in performance there too. By the time they reach the required performance level, many years will have passed and most of their so called efficiency will be due to process improvements rather than real efficiency gains. Besides AMD has much of the fixed function accelerator stuff already within the UVDx portion of the APU. Also a big disadvantage with ARM wrt x86 is the non binary portability between different ARM implementations even with the same core version. That is not a problem with most x86 systems with the possible exception of not enough memory to handle the footprint of the application (no swap available or too small, if present).