looking for explanation of why x86 cpu are inefficient?

[eksasol]

or beleived so by many members here.

Well ever since I started getting interested in linux, I learned about ARM and I'm intrigued by things like the beagleboard but it's horsepower output with such low wattage output. There are many members here who said the x86 architecture is inefficient in resource requirement. So I don't know anything about assembly languages and the engineering side of it, but for those who beleived so about the x86 architecture, I'm looking for explaination or linkage that provides educated explaination to the facts. I haven't had much results from searching on google. (I'm not looking for linux fanboy bashing similar to the replies in the OQO vs Pandora thread, inwhich some the replies were out of contex and incorrect in their comparisions.) Yes sorry I know this is a very complex issues and with so many algorithm and type of usages, which is why there are three "processor" in the Pandora. Ofcourse there are improvement in efficiency over the years, but can there be a practical change or alternative to this x86 architecture for the benefit of home and business on a world wide scale?

I want to be more broad here, not just the x86 architecture but the whole power consuming trend. I've been more concious about the consuming electricity in my daily life, the fact that my desktop running 24hours/day and calculating the yearly cost of running it gives an astonishing results (recommended to others who haven't tried it). Personally I am sick of the gaming and horsepower market fueled by conglomarete companies like Intel and Nvidia, targetting at uninformed gamers, which I really think help to inflate the cost of their cpu and gpu product values. Most humourously the newegg.com's new feature of power supply recommendation to drive up their sales. Even though I want to replace my desktop with a more energy efficience desktop, with things like microatx mobo and 45nm cpu, it is still difficult to highly reduce the wattage output with the demand for necesary computing power.

 

[mali]

You should look at a comparison between RISC and CISC, I think that is the starting point.

Edit:
braindump
I havn't used a desktop for years now. I use notebooks at home, which have a huge advantage in power consumption compared to desktops.
The main problem, as always, is the market. Cheap is more important than efficient. I think we will see some kind of convergence of devices coming from different branches. The main pushing factor for PCs has been the gaming industry in the past. Today more and more people prefer dedicated consols for gaming. Notebooks sales are growing more than desktop sales. I think the classic dinosaur-desktop will disappear. People start to weight design(which naturally incorporates more efficient components, because of heat production->loudness and big cases) more over horsepower, that isn't needed for everyday tasks. Desktops will shrink, and handhelds will get more popular, because of increase of horsepower while maintaining efficiency. The devices will meet somewhere in the middle. We will use more devices for dedicated puposes rather than on big desktop for everything, like we did in the past. Think of it as a wifi connected network instead of a single workstation. The combined power of those specialized devices exceeds the power of one desktop anytime, while being extremely efficient, because of its dedicated purposes.
/braindump

Edit2:
The efficiency differences might not be solely dependent on architectural differences(reading RISC vs. CISC implies, that the debate isn't up to time any more). I think the most important factor are simply the design decisions followed over a certain time period. The Strong ARM /OMAP design is intended to be portable, and has been developed over years now, while x86 is just starting to be developed in that direction, which is difficult, because it has to compete on one side with ARM and on the other side with more powerful(but less efficient) x86. Another very important factor is the use of dedicated chips instead of one general purpose cpu, like DSP, IVA and SGX on OMAP.

 

[eksasol]

You should look at a comparison between RISC and CISC, I think that is the starting point.

Thanks, that help explain these things a bit.

 

[Shikaku]

ARM notes:
http://en.wikipedia.org/wiki/ARM_archit ... sign_notes

Uses http://en.wikipedia.org/wiki/RISC to decrease the amount of instruction sets and therefore the amount of silicon and chips and therefore power

X86 notes:
http://en.wikipedia.org/wiki/X86_instruction_listings Read this to see how much is required to become X86 compatible. Note the http://en.wikipedia.org/wiki/X86_instru ... structions, and many other enhancements, which allow acceleration of FPU operations and other operations.

 

[anexanhume]

One of the problems is that x86 is mired in constantly being backwards compatible. They should get together and write a new ISA, but that's never gonna happen

 

[mali]

One of the problems is that x86 is mired in constantly being backwards compatible. They should get together and write a new ISA, but that's never gonna happen

Transmeta tried to go a new route with its code morphing chips, but unfortunately that wasn't honored by the consumers.
http://en.wikipedia.org/wiki/Transmeta

 

[maciek_urbanski]

One of the problems is that x86 is mired in constantly being backwards compatible. They should get together and write a new ISA, but that's never gonna happen

No it will not. And not because of bad will, but consumers just wouldn't buy a PC that does not have any software for it (maybe except assembler & compiler). Porting the abundance of x86 code to new architecture ain't going to happen...
Just-in-time re-compilation of x86 code is a solution, but it's hard - Transmeta did not manage to do 'good enough', and these guys were really good programmers...

Anyways, any modern x86-CPU (AMD, Intel, but I dont know about VIA) disassembles x86 opcodes into it's internal opcodes called uOP-s (think significantly simpler, RISC-alike instructions). So in a way most general purpose CPUs have x86->'RISC' translation and RISC core...

But CISCs are not that bad... From ISA design perspective ARM Thumb instruction set tends to gravitate toward 'CISC'-ness (limited register rangle, 2-parameter syntax). Of course it has more elegant encoding...

Like gerd wrote - power consumption was not an issue for x86 for many years. Now it starting to change rapidly... Just compare power usage of Core and Atom architectures. I wonder what Atom successor will be like...

In power efficiency arena ARM has upper hand (and, I believe, 45% of market share), but if it gets strong competition we'll all benefit from it.