What adding a dedicated GPU is not cheating? Then ARM isn't the source of the performance, the GPU is. Is that any different than a x86 with a much more powerful GPU? The Radeon 4290 (890GX IGP) is 32 times as powerful, so ARM plus GPU still has 32 times just to catch up to the IGP while Ontario leaps ahead by another 2-3 times over current IGP and Llano zooms ahead by 6 times on top of Ontario.
I guess ARM/GPU will catch up to the HD4290 IGP by 10nm, Ontario by 7nm and LLano by 2.5nm. Any of those are in the next decade 2020 or perhaps 2030 to where Ontario will be by next year and Llano 3-6 months after that.
Just say that the others were mistaken.
First it was claimed that ARM will eat into entry level PCs and Laptops. I disagreed and said that all of the ARM products were not up to entry level laptop speeds. I gave an example of one that wasn't even to the speeds of a high end laptop of more than a decade ago. Then the discussion got sidetracked on the example as to why it was so slow. Sure it was software, but that meant that ARM's speed is due to lots of careful optimization for the gadget in question. PCs and laptops are more general purpose than that and require good enough performance for anything thrown at them. And there is a lot of working crap that is forced on them.
My one example is a bad case, but that happens a lot with PC applications that are normally thrown together, work and then shipped with no thought ()or just a little) for making it optimal. 30 years ago, when cycles were precious, being optimal was worth a lot and most code on x86 was. But as x86 performance leaped and bounded higher, people got sloppy and it didn't cost them much. Once things worked, you threw hardware at it until it was fast enough. It was much cheaper than optimizing the code as only a few dozen really used it or even one. Rarely did hundreds or thousands used that application. And by the time it got distributed much wider, x86 performance caught and then passed its requirements.
So here we are where in the PC and laptop world, there is literally millions of applications for every conceivable area of use (most are one shot for that user or organization). And then there is the ARM world of gadgets where each piece of software is careful tailored for that exact gadget to be good enough and no more (more costs money, sometimes lots more). One looks at the performance of those carefully tailored applications and think this can replace most of those entry level laptops. But when that poorly built code gets on those upscaled smartphones and PDAs, their performance goes to shit and they look pitiful against x86.
Going the other way is somewhat easier. The only question is how much of that high hanging fruit used to improve x86 serial performance can be removed, reducing the size and power used and still let those poorly built applications be fast enough. Bobcat cuts half the transistors, the power by up to 10 times and still gets 90% of the K10's performance. Is that enough of a drop on size and power, yet still able to run poor applications fast enough. Given the interest, it looks like the answer is unquestionably YES!
Will it still be fast enough if they move to a low power process? If yes, power could be cut to 1% of the original, but the top side will drop by half. If almost all of those poorly performing applications still are fast enough, ARM will slowly lose the battle ground and later the rest of the gadget space. And thats because old software is dirt cheap or practically free and that difference usually is worth more than the gadget itself. And we all know what happens when that is the case.
Pete |
