The Intel Atom “Medfield” chip will finally be available in some smartphones this year, after Intel promised for at least 3 years that they would enter the smartphone market. Unfortunately for Intel, I think they are just rushing in with Medfield, because I don’t think it’s competitive enough in power consumption, and it’s quickly losing the performance game, too. That’s without even considering that x86 hardware (not a SoC yet) is still several times more expensive than an ARM SoC.
Intel might get to the same power consumption level with 22nm, but I think by then they will lose the performance game. To understand why that will happen, you need to understand how Intel, and everyone else improves their chips from year to year. Say the total improvement of either power consumption or performance for the next generation chip can be 40%.
For simplicity’s sake, their choices can be cut down to these 3:
1) They keep the power consumption the same, and improve performance by 40% – what Intel has done in the PC industry for years, keeping the same or slightly less power consumption, while increasing performance. ARM chip makers do this, too, which is why it’s pretty pointless to ask them for “better battery life”. The market still wants ever greater performance, so they will try to improve efficiency, but only so they can keep the same old TDP, while increasing performance even futher.
2) They keep the performance the same and improve power consumption by 40%. This has been Intel’s strategy with Atom over the past few years, because they needed to catch-up with ARM. They started from a TDP of 10W per core, and now they’re at around 3.5W per core.
3) They improve performance by 20%, and power consumption by 20%. This is a “compromise” strategy, where you choose to improve both power consumption and performance, but since you’re doing both in the same time, the gains are smaller for each side. This is not really a strategy Intel could choose with Atom, because they were really far behind ARM in power consumption. So they needed maximum power consumption gains.
Intel Atom has barely increased in performance since it was introduced 4-5 years ago, and because of ARM, they’ve only tried to work on power consumption, every year. They also never wanted Atom to be powerful enough to be used in lower-tier laptops, and cannibalize Core i3 and such lower end, but very powerful chips, but that’s another issue that Intel has to deal with internally.
So initially, Atom was a lot more powerful than ARM chips, but because they needed to catch-up with ARM in energy efficiency, they kept the performance more or less stagnant. Now ARM chips are about to beat Atom in performance with Cortex A15 and Krait (some tests even show Cortex A9 beating it at the same clock speed), while still having having significantly lower power consumption.
A dual core 1.5 Ghz Krait chip has a 0.75W TDP under maximum load, while Atom has 2.6w TDP in “idle mode” alone (when your phone does nothing), and 3.6W when playing a 720p video. So that’s around 4, maybe 5 times less efficient than the best ARM chip right now.
The Intel Atom Medfield chip will be paired with a PowerVR SGX540 GPU. You can probably remember this GPU because it’s the same GPU that launched with the Galaxy S in the first half of 2010. While it will be overclocked to be around twice as powerful, it’s still an old architecture in a time when even the SGX543 is one year old. And the first Medfield phones are still months away from being launched. I expect Medfield to be one of the worst chips for graphic performance in 2012, and all high-end ARM chips to have significant gains over it.
The Medfield chip also has only one core with in-order execution, just like Cortex A8, in a time when we can benefit from fully out of order chips like quad core Cortex A9 chips or dual core Krait and soon even Cortex A15 chips (thanks to the Exynos 5250). The biggest benefit of multi-core chips, besides the performance boost for multi-threaded applications like the browser and some games, is multi-tasking. Multi-tasking might not be extremely important in a smartphone, and is indeed more important on a tablet, but I’d still take at least a dual-core phone over a single core one, if given the choice.
Even if Intel moves to 22nm before ARM, ARM still has the Cortex A7 coming out within a year or so, which can be several times more energy efficient than Cortex A9 or Cortex A15, and it can be used with Cortex A15 to provide both maximum energy efficiency and maximum performance when needed in the big.Little configuration, so Intel will have one more milestone to catch up to, meaning they still won’t be able to focus too much ton increasing performance, while ARM chips makers can.