We attended the Intel Developer Forum’s (IDF) second keynote this morning (did you follow the live blog?), and the most important thing that you need to know is that Intel is making “low-power” its priority. Low-power is not a new theme: ever since the Pentium 4 got shelved, power efficiency has become increasingly important. However, the pressure from mobile platforms, the rise of non-Intel tablets and the arrival of Windows 8 on ARM-based systems has made low-power the highest priority at Intel. That is partly why an upcoming generation of Intel processors, codenamed Haswell is going to consume up to 20X less power, says Intel.
Why does power need to come down so quickly?
The short answer is “because people want it”, and Intel has recognized that during its keynote. However the more correct answer is “because people can get low-power elsewhere”. ARM-based systems like the iPad, the Galaxy Tab and others offer up to 10 hours of battery life while doing intensive work (like decoding a movie). Most PC battery numbers are based on a depletion test where the PC is ON, but doing nothing. In the real world, you would be happy if you can watch 3 hrs of video on a laptop.
How can Intel pull this off?
Optimizing power efficiency is one of the most interesting and challenging areas in computing. There are several approaches that are best when combined (but it’s very hard).
The ARM way: going low-power using Design Efficiency
Historically, ARM has had much better power efficiency because it has been designed from the ground up to be that way, instead of pursuing pure performance. Also, the ARM architecture and instruction set is -by nature- more nimble than Intel’s, but it is not backwards compatible since the beginning of time. ARM combines that nimble nature and a host of other techniques to keep power usage very low. That said, ARM doesn’t always have the latest power-savings techniques.
For example, most dual-core designs from ARM can’t use different voltage and frequency for different cores. I think that Qualcomm has added this ability to their latest chip, and I don’t know of any other company who has done it. ARM’s power-oriented design has been extremely successful, and the only (but not small!) downside is that “peak performance” may be far from what Intel has to offer, and one cannot simple “crank the voltage up” to compete. ARM licensees may dispute this, but in reality there’s no ARM-based device that actually competes in real-world performance, even with a “slow” computer like the Macbook Air 11″.
Intel: going low power by being smart
Intel may not have a lot of options when it comes down to trimming its X86 instruction set, but in the past, it has been really good at increasing performance while maintaining backwards compatibility. This time, Intel will use a combination of advanced manufacturing process, hardware and software tweaks to not only make every transistor use less power, but also to shut them down completely whenever they have a chance. As we reported earlier, Intel is also researching ways to use ultra-low (0.1-0.3V, my guess) voltage as demonstrated by having a CPU powered by a single solar panel – an incredible feat for a PC processor. This is not coming to the market anytime soon, though.
With the upcoming Haswell architecture, Intel will coordinate efforts from the operating system (OS) and the processor hardware to put (or keep) as many components to sleep mode, when they are not utilized. Intel also plans to distribute the workload to components that are already awake, instead of awakening sleeping ones.
This is a very interesting way for Intel for one reason: the transistor that consumes the least amount of power is the one that is turned OFF. Even a very efficient design can’t beat that, as zero is… zero. Obviously, ARM can try doing the same thing, but I suspect that it is not under the pressure of doing so, yet.
What does it mean for you?
ARM or Intel – why would you care? In the next 4 years, you should care because if Intel gets its way, it will let PC makers build small, light and long-lasting computers that have the functionality and speed (to some extent) of Windows machines, but near the battery life and weight of ARM tablets. Those computers will also be able to run every Windows app that you currently have, and they won’t even need to be recompiled (and re-distributed) for ARM. Every driver that works on a PC will continue to work, and today, the driver diversity is one of the greatest strength of the “WinIntel platform.
It is clear that without the pressure from low-power devices, Intel may not have moved so quickly. Competition is good for the end-user.
Is Intel in trouble?
Not really. Not yet.
It is imperative for Intel to draw a (power consumption) line in the sand at which ARM’s main added-value doesn’t matter as much anymore. If Intel fails to do that, it risks giving developers more incentives to make apps for non-Intel platforms, and applications are the lifeblood of any hardware platform. This is the reason why Intel fought so hard -and won- to preserve backwards compatibility since the i8086.
Since Windows 8 will also run on ARM, it is a gigantic protective wall that will go away. Of course, Intel stays a formidable industry player, but the X86 exclusivity of Windows was like a bullet-proof vest that made Intel near-invincible.
If Intel can continue keeping this trend, it will probably do well as long as it provides customers with what they demand: very portable, affordable, devices with which they don’t have to worry (too much) about battery life. The years ahead are going to be exciting!
Related story: Are we in an ARM Bubble?