October 16, 2013
Intel Quark Provides Spin, Charm, and Strange
New Low-end x86 MCU Line Emerging from the Lab
Intel is going subatomic in its charge to empower embedded systems.
Last week the company teased out a few hints about a new low-end product line called Quark. Quark is like Atom, but smaller. (Get it? Quark? Atom? It’s nerdy.) Like Atom, Quark chips will be x86-based embedded processors, but, unlike Atom, they’ll be more like an MCU or SoC. Indeed, the first (and so far, only) Quark chip revealed has “SoC” right in its name.
Like real quarks, the Intel version is hard to capture and pin down. The company has revealed almost nothing about Quark’s features or technical merits, preferring instead to just telegraph its intention to enter the market for “intelligent systems” that need x86 compatibility.
First, the spin. Intel claims that Quark’s beating x86 heart is one-fifth the size and one-tenth the power of an Atom CPU. Sounds impressive, but it’s just smoke and mirrors at this point. Comparing a “core” to a “CPU” leaves a lot of taxonomic wiggle room: does one include caches and peripherals while the other leaves them out? Are we comparing them fabricated in the same silicon process node? Or are we, perhaps, just waving our hands in front of the cameras? Too soon to tell, but the takeaway here is “small.”
Quark is also pitched squarely at wearable devices and the dreaded Internet of Things (IoT). Of course, you’d expect the company to say that. Every technology announcement these days includes those two terms. “Wearable” and “IoT” are the bells that trigger investors’ Pavlovian reactions. I wouldn’t read too much into the discussion, apart from the fact that Quark is intended for small-scale embedded devices. Think 32-bit MCU.
On the upside, Quark is expected to be tiny, cheap, low-power, and available in integrated SoCs like an x86-based microcontroller. It should have all the goodness of x86, but in a fun-sized package.
The initial device, called Quark SoC X1000, will run at 400 MHz and sport a DDR3 memory controller, PCIe, Ethernet, USB, and the usual assortment of serial interfaces (UART, SPI, I2C, JTAG, et al). Software support will come from Wind River, naturally, as well as from McAfee, the latter providing unspecified “integrated security” features presumably designed to thwart hacking or reverse-engineering. The chip will run Linux (though not quickly) and be compatible at some level with little Arduino boards. Given all the stuff on the chip, the x86 processor core is probably the smallest part of the device.
On the downside, Intel still isn’t licensing the Quark CPU core to other SoC developers. If you want Quark (or indeed, any x86 CPU) inside your chip, you’ll have to buy the whole chip from Intel.
The company isn’t saying much about the X1000, and even less about the broader Quark product range as a whole. There’s no way to know where the rest of this family is headed, and no guides to evaluate its roadmap. Will future Quark devices be relevant to your designs, your market, and your needs? Gonna have to wait for that. Also no word on pricing or availability. At this point, Quark is a largely theoretical particle with an unknown trajectory or half-life.
Even so, Quark certainly has its charming aspects. The x86 is one of the most popular and best-supported CPU families in the world; who wouldn’t want some of that in their own system? Low-end designers – sorry, designers of low-end systems – can get a “genuine Intel inside” processor for their Internet-enabled Coke machine or smart watch. And with McAfee and Wind River also standing behind it, you’ve got a gold-plated backup team.
Intel says the core of Quark is synthesizable, as opposed to hand-designed, which means it’s relatively easy for the company to spin out new variations. That means Quark could become Intel’s rapid-response weapon of choice for new markets and new design opportunities. We might see new Quark chips pop up faster than, say, new Atom chips have done in the past.
Quark is not without its peculiarities. Intel says it’s “based on the Pentium” architecture, which sounds harmless enough. After all, haven’t all x86 chips been based on pretty much the same architecture for 40 years now? So of course it’s based on the Pentium… right?
Not so fast. When Intel says Quark is based on Pentium, they really mean it’s based on Pentium. As in, a 20-year-old chip. As in, rewind the clock and undo all the changes Intel has made to its processors since 1993. No MMX media extensions. No streaming SIMD (SSE) instructions. No out-of-order execution. No 64-bit registers, instructions, or addressing. In short, none of the stuff that you probably grew up expecting to find in a modern x86 processor. Talk to your parents about what Pentium processors were like back in the day and you’ll get the idea.
Quark seems deliberately hobbled, as if to protect Atom. There’s nothing to prevent Intel from producing an MCU-friendly x86 core with more modern features in it. Well, nothing technical, anyway. The barriers here are purely self-imposed. Exhuming a 20-year-old Pentium design seems like a remarkably big step backwards for little apparent benefit. Programming tools – one of the strong points of the x86 ecosystem – will have to be rejiggered for the retrograde motion of Quark. Time to blow the dust off your old 5.25-inch floppies and boot up the DOS-based development system. Does Windows 3.1 run on this thing?
Top and Bottom
With Quark, Intel has filled out the periodic table, from heavy metal (Xeon) to the lighter elements, such as Atom. About the only thing the company doesn’t offer now is a 16-bit MCU based on the original 8080 or 8086. Quark will – or at least, is intended to – appeal to designers at the bottom end of the range for whom Atom is too big, too expensive, or too fast.
I wouldn’t expect quark to be cheap, however. Intel has a reputation for making premium-priced processors, and Quark is likely to follow suit. If the initial X1000 is any indication, Quark chips won’t be stripped-down generic MCUs. They’ll be heavily loaded with peripherals in order to justify Intel-level prices. But x86 compatibility and Intel-quality silicon don’t come cheap. If that’s what you want for your next wearable Coke machine, then the solution is elementary.