VMWare CEO: Intel's x86 filled with Junk silicon

[Modelworks]

Originally posted by: IntelUser2000

Originally posted by: Modelworks

The new ARM chips are multicore and include a dsp that can process HD video + interfaces + memory and still use less power than the Atom cpu alone. Add in support chips + memory + video and the Atom has a long way to go. The Atom cpu will have to run on lower power itself before it could compare, just making it a SoC design will not solve that problem.

Straight from the horse's mouth, Cortex A8 is claimed to use <0.59mW/MHz without cache memory. At 800MHz, that is <472mW. Since its less than, lets assume 450mW.

Atom Z500 at 800MHz is at 0.6W, or 600mW. The difference is that Intel includes everything including L2 cache unlike ARM.

CPU-wise, they are very comparable. Second generation Moorestown will probably be using more architectural tweaks and their specialized low power process to bring that down.

Before Atom, people didn't even think about x86 in the phone place. Now there are two sides, one supporting the theory Atom will get there, one that doesn't.

What will it be with Moorestown?

Except you are comparing them clock for clock. ARM doesn't need a equal clock speed to do the same task as a x86 cpu. A 600mhz ARM can browse and run java apps equal to a 1.6Ghz Atom, which uses 2Watts just for the cpu. The ARM uses 1.6W for everything including the video + ram+ interface. And that was before the upcoming shrink to .13u.

The problem Intel still faces is they are stuck having to keep it x86 compatible. ARM doesn't have to do anything but make it run faster. If they have to add or remove instructions or registers, they can do that. Change the compiler for that chip release and they are ready to go. The apps will be written with the new changes in mind.


What would be a smarter move for intel is to drop x86 for the embedded market. Sit down and figure out a processor from scratch that will benefit that market the most and not have their hands tied with the x86 way of doing things. Then put all that engineering ability into a product that it totally focused on the market.

A general purpose cpu was great for a time, but I'm seeing some truly incredible special purpose processors come on the market that nothing in x86 can touch. Especially in the dsp field. It is like how we made the transition from cpu graphics to gpu graphics. Sure the cpu can do graphics but a specialized processor has the upper hand.

The other thing Intel faces is that it isn't only one company they have to outdo in engineering. It is all the ones that are ARM partners. That list is pretty impressive:
Samsung, STMicroelectronics, Texas Instruments (TI), Broadcom, PMC-Sierra, Matsushita (Panasonic) and Freescale. Those signed up to Cortex-A9 include NEC, Nvidia, STMicroelectronics, TI and Toshiba.

 

[IntelUser2000]

Originally posted by: Modelworks

Except you are comparing them clock for clock. ARM doesn't need a equal clock speed to do the same task as a x86 cpu. A 600mhz ARM can browse and run java apps equal to a 1.6Ghz Atom, which uses 2Watts just for the cpu. The ARM uses 1.6W for everything including the video + ram+ interface. And that was before the upcoming shrink to .13u.

I so have to disagree with this. Have you seen the Iphone 3G and 3GS review?

http://www.anandtech.com/gadge...howdoc.aspx?i=3595&p=6

The Atom devices get sub-10 second loading times. I told you earlier, clock per clock, the Atom is 40% faster than the Cortex A8.

The times where instruction set matters for performance is over. Implementation and process technology is much more important.