INTEL DROPPED A bombshell on the market today with a little bullet point and a few carefully chosen words. Opening up Atom to third parties to interface with willy-nilly went from impossible to product in the space of one slide.
During the morning keynote, Intel Senior VP Anand Chandrasekher, started off with the usual internet on everything theme that Intel is so keen on. Atom is aimed directly at this, but Intel swore early on that the chips would be the anti-ARM, IE generalist and closed.
When ARM refused to roll over and die, then politely decompose, Intel had a fight on it’s hands. The initial battle was framed as Intel’s superior horsepower and horsepower per watt vs ARM’s task specific silicon and accelerators. Would Intel’s silicon advantage overcome the plethora of ARM devices aimed at every niche under the sun? Will the x86 decode overhead bite Intel at the lower end? Two strong oxen or 1024 chickens to plow the field?
For the past few months, there were whispers that Intel was looking to do some foundry business. Feelers were being put out, and expectations gaged, but not much more. On the flip side, Intel was scouring the Valley for any silicon IP that might work well with Atom CPUs. Nothing definitive, just looking. If asked, no external IP was going into Atom. Ever. Really, they mean it.
The first crack was Intel allowing TSMC to fab task specific Atom variants for their customers. Intel letting it’s cores out? Who would have thought they would ever do that? Likely anyone who realized that the alternatives were letting their process tech out or ceding markets to ARM. The door was cracked open.
So, Intel let a smaller crown jewel out of the fold, but kept the big ones, IE their process tech. Today, Intel threw proverbial barn doors open when Anand mentioned that all the new Intel parts, Atom and Nehalem, were modular. Hmmm…. Then he said that on Intel’s 32nm process, they would be opening up the on-chip bus. Blink. What? Really?
Change looks like this
Notice the line that says “Shared, modular library at 32nm”? And the part a few lines up about derivatives? Well, the whole thought process of performance per watt vs task specific silicon was just won by task specific silicon. The ARM thought process is now the Intel thought process, plus Intel has the horsepower that ARM doesn’t. Yet. Then again, ARM has the lower wattage space nailed down, but Intel is heading that way fast.
So, with a new, more open and friendly Intel, the question is, does Intel drive power down to phone use levels before ARM gets the drivers and Firefox plugins that x86 has? The race is on, but another hand grenade was tossed into that ring by Google last week when they announced Chrome OS.
MS is surrounded by enemies picking at them, one of which is Intel. Intel is surrounded by enemies, picking at them, one of which is MS. Neither likes ARM much, but Google’s apps run pretty well on ARM. I see a few alliances forming. Talk about being nibbled to death by ducks.
Intel had to change, and did. I for one think it will work out very well for them, much better than the way of Intel in decades past. Change may be unpleasant, but it is better than the world going around you, just ask the record companies.
In the end, there are two more open questions. How open is Intel really? Is this a toe in the water, or a 180 degree turn? Will it last? Do you want to bet your company on that? The other big question is will Intel fab Atom++ SoCs (System on a Chip) for customers at their fabs? Do they have the capacity? Will that mean process information leaks, and do they want to let that information out? If so, what happens to TSMC?
With that one bullet point, a few hand grenades going in and out, and some background knowledge, Intel changed the game and their company. It is going to get interesting from here, Intel has one heck of a balancing act ahead of them.S|A
Latest posts by Charlie Demerjian (see all)
- Globalfoundries 7nm process isn’t even close to the name - Sep 26, 2016
- ARM upgrades realtime offerings to v8-R and adds Cortex-R52 - Sep 21, 2016
- Everspin and Globalfoundries team up for embedded ST-MRAM - Sep 15, 2016
- Intel’s Xpoint is pretty much broken - Sep 12, 2016
- ARM adds 2048-bit vectors to v8A with SVE - Sep 7, 2016