BUFFALO, N.Y. Recent press coverage of the Intel Atom got much of the story wrong.
Atom is a new architecture that offers Pentium M-class performance within a 2-W power envelope packed into a 25-mm2 die. These numbers are very small for an x86 processor, and they allow Atom to target embedded applications like mobile Internet devices where the x86 has not been used before.
If you believe the buzz, Atom is so good that it will displace ARM cores as king of the mobile world. Unfortunately, the buzz was extremely confused. Here's what the coverage got wrong, what it got right and what it completely missed.
What they got wrong:
Atom will beat ARM because it can run Vista. "They've got to be kidding," said Will Strauss, president of research firm Forward Concepts. Mobile devices will not run Vista. "It's a memory hog and power-consuming hog. Maybe Windows CE." Linux is an attractive option for these devices. ARM can run Windows CE and Linux just fine, so Intel has no real advantage.
Only Atom offers a "real" Internet experience with Flash video, YouTube, etc. Wrong. You can buy ARM Flash players from vendors like BSquare. On2 just announced a YouTube video decoder for ARM. "The Internet doesn't care about the architecture or the operating system," said Jim McGregor, principle analyst at In-Stat. "It creates a level playing field."
Intel dominates every market it enters. This is silly. Intel already tried to dominate the mobile market with its PXA line, and that effort failed. It ended up selling the processor line to Marvell.
Atom will win because ARM is proprietary technology. More nonsense. You can get ARM chips from Freescale, Texas Instruments, Qualcomm and others.
Intel will win on cost. Intel hasn't announced pricing, but it is obvious that Atom will cost far more than ARM-based chips. An ARM Cortex-A8 occupies less than 3 mm2 using a 65-nm process. Atom weighs in at 25 mm2 using a 45-nm process. In fairness to Atom, most of its area is occupied by cache. My analysis suggests that the Atom core itself is about 9 mm2, or three times bigger the size of Cortex-A8. With such a huge area disadvantage, it's hard to see how Intel will win on cost.
Intel will win on power. "Intel admits that the current solution is not as power efficient as the potential ARM competition," said analyst Strauss. Intel quotes thermal design power of 0.6 to 2.0 W for Atom, but doesn't specify the corresponding clock speeds. Comparing this to ARM is a bit tricky, since ARM uses typical power numbers. (TDP is a maximum.) Let's generously assume that Atom's typical power is 50 percent of TDP, as is the case with the AMD Geode LX. This gives us about 300 mW at 500 MHz. ARM quotes 0.59 mW/MHz for the Cortex-A8, which also corresponds to 300 mW at 500 MHz. Thus, the best-case scenario is that Intel matches ARM. In most scenarios, Atom burns more power.
Great article. Artcile mentions that Atom cannot run vista.
But some where I saw that atom can run vista? See this product
I don't think porting to ARM is the largest bit of work - instead, what takes time is porting to new APIs. And I'm not just thinking of OpenGL ES here, but also the OS's APIs.
Either way, I don't think it matters: since when do handhelds have keyboards or mouses? :) The control scheme and the lower level of performance are even bigger problems than APIs, I suspect.
Also, can you confirm with 100% certainty that the Centrino Atom, which will be used in desktop and notebooks, uses a SGX core? And if so, do you know which core it is? My understanding is that it'd have to be one of SGX's higher-end cores (potentially with DX10 support), and I suspect that's not really appropriate for handhelds...
Arun AT beyond3d DOT com
"Atom can run PC games; ARM can't."
This makes no sense! Games are not ran on the CPU, but on the GPU. Even taking this a bit farther, it would be more appropriate to say that they are ran on the 3D stacks like Direct3D or OpenGl ES rather than CPU or GPU.
Cross compiling a game for a specific instruction set is not a bottleneck in the game industry - almost all ISAs have efficient compilers and assemblers (of course we can ignore failed ISAs like TI VLIW etc...)
Major concern is providing game experience in terms of visual effects, power and input handling ? for all those things a CPU will not matter. A GPU matters.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.