Opinion: Intel is Underestimating ARM

[meltbox360]

[citation][nom]RipperjackAU[/nom][citation]If you are gaming on a laptop, (NOTE: YOU ARE DOING IT WRONG!) then sure, go ahead.[/citation]Funny thing you said that, because RAGE runs better on my nVidia equiped laptop, then it does with my ATi/AMD CrossFire gaming rig![/citation]

Meaning that is you had a ati laptop it would run better than on a Nvidia desktop right? You really just needed to prove him wrong eh? Anyways Rage heavily favors Nvidia because ATI openGL drivers aren't that good... at all. Take DirectX games and then its pretty much a tie. Games that choke GPUs tend to play better on AMD cards this generation due to the large ram and bandwidth. CPU limited games play better on Nvidia. Nvidia has had lower cpu overhead for a while now. Each has strengths. Both have driver problems and both have games that favor a card which Nvidia usually had the advantage in but recently strange stuff had been happening Dirt is now favoring Nvidia cards and Crysis is favoring ATi cards.

Anyway, that's my heavily off topic post!

 

[meltbox360]

[citation][nom]Anonymous[/nom]Yeah... right...Here we have AMD, who now provides 80% to 110% of Intel's performance in everything except synthetics that were compiled with ICC, and they're "slow, uncompetitive, dissapointing", and generally considered not suitable for anybody's PC, since we all need soooooo much processing power all the time. Now we're suggesting the real battle is smartphones and tablets, even though Intel is not remotely competitive, and never has been, we're suggesting they're somehow the incumbent, and will now have to fight off ARM for taking their non-existent piece of the pie?This part is just classic:""As the need for computing performance goes up, both the Intel architecture and the ARM architectures face the same fundamental physics problems, which is more performance requires more transistors"Orly? Is that why the Atom chips for smartphones, etc... are bigger than ARM's entire SOC just for the CPU, consume 4x as much power, and are nowhere near 4x as fast? Yeah, you and ARM are both just fighting the laws of physics now, right....[/citation]

I agree with parts of what you are saying. AMD is behind Intel in raw performance no doubt however I suspect Intel gains a huge advantage overall from simply having a larger team working on compilers. AMD does not have the resources to do that and so they suffer. The other thing is that Intel also tends to be the targeted platform if someone takes the time to do some optimization due to the fact that they are the majority. No matter all that AMD is still behind at least in floating point performance and probably a little bit in integer performance.

Now to address ARM. Let me just point out the Intel had experience building incredibly complex chips with billions of transistors. High performance chips are hellishly more complicated than low performance chips.

Here is an example:
If I tell you to go out and build a remote control car that goes 10mph you may be able to do it. If I tell you to build a remote control car that goes 130mph things get a lot more tricky. Parts have to be more durable etc.

Of course cpus are different but in high complexity cpus there is a lot of interaction between various parts of a cpu core. Everything has to work perfectly. In low performance cpus a lot of the complexity is lost. Since Intel is moving into low performance low complexity cpus they will have to learn the tricks that are employed in low power but they already know how to squeeze performance out of incredibly complex systems so doing the same with less complex systems will be that much easier. Just seems like it is a matter of time.

The comment on how they are battling physics.... Right so each GPU generation defeats one of the laws of physics allowing Nvidia and ATi to continuously improve performance? Brilliant!

But actually they just figure out how to bring transistors closer together packing more on a die. So yes they fight physics but that does not account for all of the performance increases we have seen. The key is figuring out the most efficient layout given your process limitations and transistor limitations (power/die size).