THE BIGGEST LOSER in the tie up between Microsoft and Nokia is none other than Intel. While they may try to put a brave face on the matter, the simple fact remains that their Atom line is now without a future.
As you may recall, Nokia was going to hold up the flag for Intel and their Atom chips in phones, starting with Moorestown, the 45nm Atom SoC variant, and moving on to custom silicon at 32nm. That variant, as we exclusively reported, was called Penwell. Nokia’s deal with Microsoft just killed that chip dead, and any hopes Atom had in phones went with it. Meego was hit in the head with an errant bag of cash during the drive-by, and has family gathering in the emergency room to pay their last regards.
Technically, the Atom core is doing everything right, each variant hitting internal targets and improving at a very quick pace. Moorestown was a hugely impressive chip, able to kick the highest end ARM variants to the curb with it’s raw CPU performance. When it came out, nothing was in it’s league, and while ARM variants came close months later, the next iteration, 32nm Medfield, is due out in the near future.
Medfield will raise the bar, it’s just that no one will care. Intel could come out with the greatest silicon ever, but it will have the one stigma that the company can not overcome, x86. At this point, Intel’s greatest strength is now it’s greatest weakness. All x86 brings to phones is a decode power penalty and a lack of software, no one wants an x86 phone any more. Even Microsoft has cast their mobile lot in with ARM, something quite unthinkable a year ago.
This lack of caring comes from two distinct factors, features and software. They are both Intel choices, and mirror the company’s vision, or lack thereof. In anything that does not involve the processor core or the semiconductor process used to make the chips, Intel is not a player.
In the case of the core and the process, Intel has no peers, period. Since the P4 debacle, Intel has had a world class core that no one can touch, generations ahead of the competition. They beat AMD cores in just about every objective test, and the ones that they didn’t win at fell to the next generation chip. By the time Nehalem came out, AMD had no core that could compete, or even come close.
The same holds true for the manufacturing side, Intel has been shipping 32nm parts for over a year, and is going to ship 22nm before the end of 2011. AMD has yet to ship a single 32nm part or a single High-K/Metal Gate chip. By the time AMD gets those technologies up to a volume ramp, Intel will be months away from their 22nm 3rd generation HKMG launch. Intel is similarly ahead on advanced packaging technologies.
With such a world-class line up, why is Atom dead? The core is absolutely world class, it may not be as fast as the big x86 chips, but it was never meant to be. Intel’s 45nm LP process is similarly in a class of it’s own, and the upcoming 32nm LP process should be just as far ahead of the competition as 45LP was. What the problem?
The first problem is the ‘uncore’, aka everything that is not the CPU core itself. With Atom, Intel has utterly dropped the ball on graphics, one of the most important parts in this class of chip. It is arguably more important here to have a good GPU, or at least a passable one, than on the bigger CPUs. Why? Because you can’t put an add-in card in a System on a Chip (SoC).
To make matters worse, Intel is the sole provider of Atoms, you get what they provide, the CPU equivalent of, “any color you like as long at it is black”. You can’t go shopping for a Moorestown core with a different GPU or accellerator, Intel provides it, and you take it. ARM on the other hand has dozens of companies that will sell you just about any variant you could dream of, or make you a custom chip if there is nothing on the parts list that fits you like a glove.
With Atom, the GPUs Intel provided have been awful. Worse yet, necessary extras like a hardware video encode/decode were sometimes simply external chips tacked on to the PCIe bus. The ‘PC way’ of doing it in software had an untenable power cost in the phone and widget space. Atom was always a day late and a dollar short.
In addition to that, Atom was pushed to be the best thing since sliced bread in the widget space long before it had the power usage characteristics to play there. The first Atoms sucked power at a prodigious rate, and any device that used the chip, while not lacking in performance, could not be far from an AC outlet for long. No one wants a fast device that can’t stay powered on. Later chips mostly corrected this problem, and the current parts are simply world class from a platform power perspective, but still lack features. The early Atoms were pushed into places they did not belong, crashed and burned. By the time the feature set caught up, the world had moved on, and Atom was behind again. No one cared about the one ‘unique’ feature, x86, they had long since moved on.
Netbooks are a similar story. If anyone has tried to use a netbook with an Atom in it, the chips were promised to bring the ‘full’ internet experience to the low power world in a way that ARM products could not. They did. The problem is that if you can run everything that a big PC does, most users want to do what they do on a full PC. Atom could not deliver enough horsepower, and Microsoft’s crippled versions of Windows didn’t help either.
Atom is simply too slow to play with the big boys, and all variants since the first one did not up performance, they just dropped power by huge amounts. This solved the widget/SoC side’s greatest problem, but it similarly doomed the netbook side.
Intel left AMD a hole in the market big enough to drive a truck through, and at CES, AMD drove through with Brazos/Bobcat. If anyone buys an Atom based netbook in the coming year, you can be pretty sure they either work for Intel or don’t realize what they are buying, it is that bad. Atom is in a no-mans land for performance, features, and timing, but has nothing to offer customers in place of those holes.
So Intel has a brilliant core that is aimed at the wrong places, was pushed into spaces that it should not have gone, and got the needed optimizations too late. The uncore, especially the graphics were similarly inadequate, and haven’t seen anywhere near the improvement that the core has. Woeful is not too strong a term here.
The other major problem facing Atom is software, especially drivers. Intel has a track record in making graphics drivers that are simply unmatched in the world of chipmaking. I can honestly say that I can not think of a single company that has made drivers as badly as Intel for as long, even if you take the awful hardware into account. Even a blind dog occasionally finds a bone, unless they work for Intel making graphics drivers.
Modern Intel graphics are all based on an architecture that debuted in the 965G chipset many years ago. It was followed up by the G35, G45, G55/Westmere (the current i3/5/7 line), and now Sandy Bridge (i3/5/7 2xxx chips). Five generations in from the 2006 launch and they have yet to release a single part that met the most basic of expectations.
How bad are they? 965G literally never got working drivers for it’s entire life. There was one variant late in the game that sort of had functioning drivers, but everyone who bought the 965G never got the DX10 chip they were promised, and many DX9 features were simply broken. The G35 almost had the same fate, with mostly functional DX10 drivers released scant weeks before the G45 was released.
The G45, aside from broken HD decode in the first steppings, had somewhat working DX10 drivers, although if you tried to use them for actual gaming, you might get a lot of unintended visual extras. Or visual voids. Or crashes. That GPU came close to working, but never hit the mark. At least it had drivers from near the beginning, you can call that progress.
The current Westmere chips, aka non-2xxx Core isomethingmeaningless, finally got to mostly functional DX10 months after AMD went to DX11. Early indications that the G55-class GPUs in these parts ‘would not suck’ proved sadly wrong, and drivers were the biggest part of that debacle. This one raised the bar to ‘nearly functional’, but that was about the best Intel did.
Current Core isomethingmeaningless 2xxx chips advance the state of the art to DX10.1 over a year after the rest of the world has moved to DX11. Unfortunately, these parts will never get DX11 drivers, the hardware is incapable of supporting it. Maybe next generation, but that brings along a new driver architecture too, not something Intel is good at supporting.
The less said about the current Sandy Bridge parts the better, minor things like OpenGL just don’t work even on Windows, and what good are frilly features if the necessary basics don’t actually function?. Intel graphics drivers have gone from pathetic to simply inexcusable, I cannot understand how the company could be this bad for this long.
What does this have to do with Atom? Easy, it sets the stage. Intel has a 5 year history of failing to deliver on even the most meager of graphics promises. Things that are users take for granted as baseline functionality elsewhere just do not work and are simply never fixed. Sadly, on Atom, the first parts that Intel delivered had the same fate.
If you recall the Silverthorne family of GPUs had been offered on several products, most notably the Dell Mini 12, with Ubuntu Linux. Intel delivered video drivers that didn’t work, were never fixed, and everyone who bought the parts simply got burnt. Intel is either unapologetic or totally silent about the affair depending on who you are talking too. They used their customers and abandoned them in a cloud of tepid excuses. Buyers, and likely OEMs, never got the help they needed.
While things have gotten better, burning the Linux community in an area that is the purported target for your chips is a bad idea. If Windows is far too bloated to work in that same space, playing the hand that Intel did is a fatal mistake. Simply put, Intel destroyed Atom’s few remaining opportunities in the widget and phone world through these actions, along with their own reputation.
No one wants Windows in those markets, a point that has been made very clear by the ‘success’ of Windows Phone 7. How many billions did it cost MS to ‘partner’ with Nokia again? Intel had a golden opportunity to kick ARM out of the high end device market, but instead they missed the chance. With Intel proving unequivocally to the Linux community that they were not to be trusted, it was game over for Atom. Any good will that Intel had earned was vaporized, and then salt rubbed in the open wounds. Atom was dead in this space, and if you had to point the finger at one specific thing, it would be graphics drivers.
Since then, Intel has improved a great deal in some areas, platform performance per Watt being the big one, and Atom is world class there. Unfortunately, things like the GPU itself, and the drivers necessary to use them, while improving, are getting left in the dust by ARM and it’s partners. If you were starting a project, would you use Atom given that history?
The ARM A9 cores, especially the dual core chips, are now close enough to Intel in raw CPU power to be a wash for most people. On everything else, ARM wins, and knowing what I do about the roadmaps on both sides, Intel has no chance of catching up. To make matters worse, ARM has functional drivers, Intel does not. ARM has somewhat open drivers, Intel blames buyers and repeatedly slaps the Linux community in the face.
At this point, Intel’s Atom seems quite dead, there is no point in the company wasting more money on the line. There are no products of note using the chip, and there have been almost as many pulled design wins as Nvidia’s Tegra 2. Atom is caught between dozens of ARM cores below, many of which are better and cheaper than Intel’s products, and AMDs Bobcat above. In either market, the hardware is outclassed at key features, and neither opponent bears the software stigma that Intel does.
I thought Atom was going to change the game, it showed signs of being a world class product, but instead missed the mark every time. Intel is now in a position where it’s two strengths, unbeatable CPU cores and unmatched process tech, won’t help, and x86 is an impediment. In the areas where they are weak, uncore, especially graphics, and drivers, it is painfully clear that Intel is incapable of fixing their problems. It is time to pack it in, Atom is dead.S|A
Latest posts by Charlie Demerjian (see all)
- Thing go bump(gate) in the night for Nvidia’s GP100 Pascal GPU - May 3, 2016
- Cavium’s Octeon TX blends compute and packet moving - May 2, 2016
- Qualcomm releases Zeroth API to developers - May 2, 2016
- Another detail about Qualcomm server SoCs revealed - Apr 27, 2016
- AMD finally really honestly launches the dual Fury - Apr 26, 2016