In a completely unfair test, the folks at Tweakers.net spent a few minutes running benchmarks on an Android tablet prototype with an Intel Atom Oak Trail chip. Intel started showing off early models of these tablets at the Computex trade show in Taiwan this week, but they’re nowhere near ready for prime time and the software hasn’t really been tweaked well to run on the hardware, so it’s hard to do an apples to apples comparison and say that the benchmark scores really say anything about the relative merits of the Atom chip or GMA600 graphics over NVIDIA Tegra 2 and other ARM-based solutions.
But once you acknowledge the fact that the tests only tell a partial story because we’re looking at early prototypes, it becomes clear that Intel still has a lot of work to do.
The Compal tablet also notched a pretty lousy score on the Linpack benchmark. It actually held its own in Quadrant, with a score of 1978. That’s not great, but it’s also not awful.
Does this mean that Intel Atom-powered Android tablets will be sluggish when they come to market? Not necessarily. But the low benchmark scores do help explain why the tablets on display at Computex this week have seemed kind of slow and unresponsive.
I’m going to guess that the Dalvik engine can be further optimized for Atom. It’s based on Java so it runs on a VM and hence can run with little modification on most modern computing architectures, however the VMs can be optimized for hardware, and to expose hardware capabilities through the abstraction layer. So it’s highly likely that this is the minimized execution speed we’ll see on this hardware and, will likely get faster.
The question will be whether the x86 CISC ISA that Atom uses brings anything to the table in this arena. It’s very possible that Intel is going to need to build a first class compiler for Android that makes it easy to optimize code for CISC over RISC execution and then convince everyone to use it, which puts them in a interesting position. RISC based ARM processors are already in use pretty much universally, and if Atom isn’t signficantly more powerful, there’s little reason to move to it. If you need to specially compile code to execute on a x86 branch of Android to leverage the Atom’s supposed architectural advantages… Well the sheer mass of inertia should keep them from ever penetrating the market since they’ll have the classic chicken and the egg issue that every new platform competing against iOS and Android has.
This is assuming that straight VM optimization won’t make Atom 2x-3x more powerful than ARM’s Cortex A9 processors, which is what’s going to have to happen. You’d need that kind of clock to clock performance to justify the hit you’d be taking on the power front since the last time I checked Atom uses something like 5x as much power. This isn’t an issue if it can rush to wait quickly enough that functionally their is no power difference under a standard work load, although the processor would drain a battery much faster under the load of something computationally demanding, and games would be horrific since they’d light up the single biggest drains on the battery: the screen and the GPU, plus a more hungry CPU screen making it the perfect trifecta of power sucking glory.
All in all this isn’t good news for Intel. To my mind they needed to have a better product when it first went out to the public than this. They’re already way late to the party, having the first prime time exposure of Android on x86 be this slow is VERY bad PR to my mind, because now they are facing questions about performance. The battle was going to be hard enough since they are facing a multitude of opponents, each using chips that appear to have more clock headroom, much better performance per watt characteristics, and allow for multiple vendors for the part.
In all I think this shows that the Win8 demos on ARM scared them into releasing a technical demonstration too early, and they pretty much shot themselves in the foot by doing that, because now they’re going to face questions about A9 vs. Atom on Windows too.
So good job Intel for getting Android to run on your chips, something absolutely NO ONE thought was going to be impossible to do. Not so good on not optimizing the code to make your chip look like a energy hog who can’t even compete with the last generation of ARM chip (assuming that ARM delivers on the A15 like they seem to be doing).
Intel’s only hope is that the process shrink to 22 then 14nm will give them the transistor budget to compete per watt and performance per clock, because otherwise it might get ugly fast.
Oak Trail is still last gen ATOM, in terms of technology, they won’t go 32nm until Clover View replaces Oak Trail.
While besides the 3D Transistor technology, the 22nm Silvermount is when Intel will completely re-work the ATOM line. We’ll see things like out of order processing (all ATOM are presently In Order Proceesors) and up to quad cores introduced. So we’ll see then whether the ATOM has a future against ARM or not…
Yep. And by then we’ll be dealing with 32nm A15 ARM cores will finally have 64bit support… So we’ll see. It’ll be an interesting time.
I personally think that Intel has been far too conservative with Atom. I know I’m not alone in that, but I think it’s becoming clear that they didn’t expect ARM to be able to ramp up like this with out of order low power cores with 2ghz or more of clock headroom at under a watt. This kind of makes sense given how long ago the Atom core was designed and how little refreshing it’s had since the core was initially released at least as far as performance optimization goes.
You’re right. Silvermount will be telling. That’s when Intel will come out swinging. However by then we’ll have Win8 running on ARM, which won’t bring true legacy software support, but certainly .NET legacy code will run without recompiling unless it does something exotic… Then we’ll have ARM A15 cores hitting the market, and Denver around the corner…
It’s exciting times, It’s been over a decade since I’ve seen this much interesting development in the CPU space, with parallel chip structures duking it out. I never thought I’d see RISC and CISC dueling it out again, but it looks like we’re about to go there yet again.
Should be fun. One thing is certain though, Intel can’t just sit back any more. If they have to wait until the end of next year, and early 2013 to enter the market in a serious way because the current Atom core can’t compete head to head against A9s running Android… Well given device lead times, it’s going to be interesting to see if they can dig themselves out of the market share hole they’ve dug themselves in the mobile space.
This all has a potentially large impact on the desktop side of things too, since from what I’ve seen more and more corporations are considering moving to the older client/server model where all they need to throw on a employees desk is a thin client, and ARM cores are rapidly reaching the point where they can compete in that arena at very low margins. And if they make headroom there, Intel is in for a fight… And without the money coming in from a huge base of consumer and business systems generating revenue for research they might have issues going forward.
We’ll see… Intel could also knock it out of the park with silvermount and latter generations like they have since Core 2 came out. Only time will tell.
I agree about Intel being far too conservative with the ATOM line but that’s their tendency when not facing active competition. Fortunately they’re finally facing competition and whoever wins out it’s win win for us consumers who want more from our mobile devices.
ATOM fails again, what else is new…
Comments are closed.