Intel ULP SoCs

I've just purchased a cheap Chinese tablet containing Baytrail-T. Ordered from an Aliexpress seller.

I paid £125 for a device with a 9.7" retina display, 2GB memory/32GB storage (plus a microSD slot). It also has SIM slot for 3G HSPA data. Reported battery life is 5 to 7 hours. Almost exactly the same dimensions as the iPad Air. A lot of hardware for the price, so you can see that Intel must be practically giving the Baytrail chips away. Also, there are a lot of these sorts of devices being released:

http://mikecanex.wordpress.com/2014/11/12/bam-windows-8-1-tablet-for-99-in-u-s/

A huge amount of processing power and great potential utility for a tiny amount of money and there are lots and lots of these 1GB/16GB devices around, some of better quality than others, obviously.

My tablet (which I've barely had chance to look at since it arrived earlier this week!) comes with Android 4.4 installed but this apparently isn't the best optimised OS installation and there are reports that Windows 8.1 runs much more smoothly. I do plan to install Windows in due course but I was a little disappointed to discover that the Windows 'Bing' release doesn't work properly on this tablet (it is an unlicensed beta release), so I'm going to have to fork out some cash for a Win 8.1 license.

Even after this expense is taken into account, it will still be a very cheap and extremely good-value device and I'll be happy enough if it lasts a few years as it feels well-built.

 

That's using the worst forecast available from Morgan Stanley. As the article and Morgan Stanley themselves note, most other analysts have more bullish forecasts.

But regardless of the actual amount forecast, everyone is forecasting a loss for the next year or two. Although some (Intel) forecast breaking even by 2016. IMO, the only way that happens is if Windows tablets take off in the 99 USD space and then drives demand for higher end Windows tablets. Unlikely to happen in a large enough way in the next two years, IMO. But I'm bullish on it eventually happening, and Wintel devices eventually pushing Arm+Android devices into a niche corner if their progress continues. If that happens they'll be able to recoup lossing in the low end via higher margins in the high end.

With Broadwell already being integrated into fanless tablet designs that run cooler than fanless Tegra designs, it's going to get harder and harder for Arm SOCs to maintain better power/performance in the low end meaning that eventually their only competitive advantage will be in price. It's going to be interesting to see where the follow up to Baytrail goes considering Broadwell is encroaching on its power envelope.

Also note, that the problem of profitability in the low end tablet space isn't limited to Intel. Qualcomm is going to be finding it increasingly difficult to compete and remain profitable in the face of numerous cheap Chinese manufacturers. They've already had a few forecast adjustments putting their revenue forecasts below analysts expectations due to this combined with growth moving away from high end smartphones to low end smartphones. Which happens to play right into the hands of the Chinese manufacturers.

Interesting to note that while Intel stock has risen 30%, Qualcomm stock has fallen 11%. Showing the relative faith analysts and stockholders have in the direction each is going. Note that for Intel this obviously isn't a reflection of faith in just the mobile unit, but Intel as a whole of which the Mobile unit is still just a small, but growing piece.

Regards,
SB

 

Indeed, the fanless Broadwell is doing insane things in the ULP space, and I too am now very curious as to how this foreshadows Baytrail's successor. Three years ago, would anyone have believed that Intel would push that kind of x86-64 compute into a <5w power envelope?

 

Shouldn't the question be: what took them so long?
If you start from the idea that a high-performance ARM doesn't have many intrinsic perf/W advantages over x86, and that Intel has the best process, I thought it was just a matter of time for Intel to get its technology act together.

Which doesn't mean that I'm not impressed by the performance improvements of SOCs in general since, say, the first iPhone. My still very functional first-gen MacBook Air 11 gets blown away the chip in my phone.

Now the question is: what do we do with all that power?

 

My guess is that they woke up around 2010-2011, just after ipad killed the netbooks and the PC market started tanking. The rest, is just the usual chip design latency.

 

Windows 10 looks to be doing a good job of stopping bloat and it may actually end up reducing the required specs.

So the next baytrail can simply keep similar performance if it gets to sub 1w TDP right now its 5 to 10w compared to its haswel counter parts that were 15w TDP . Broadwell is down to 4.5w TDP while loosing only about 10% performance . So I think sub 1w is possible with baytrail performance.

I think that would make a killer tablet .

 

They'd have to get people to stop buying Android tablets and buy Windows tablets instead. Because the makers of Android tablets can use Intel or one of several ARM SOCs.

The big volume in tablets is supposedly from cheap Chinese tablets with brand names unknown outside of China. We're talking about things that cost under $100. Intel can't make money at those prices. Presumably to lose that kind of money, they dumped a lot of product below cost or got stuck with a lot of inventory they couldn't sell.

Or both.

If they dumped chips below cost, it's not going to help them in the future because if Intel tries to raise prices, the device manufacturers have several ARM SOC vendors to choose from.

Even if they managed to achieve a big performance advantage, the only devices using the fastest mobile chips are Apple and Samsung, which both make their own SOCs or buy from Qualcomm in the case of Samsung. And the below $100 market doesn't care enough about performance to pay the kind of prices Intel wants.

Intel could keep at it but at this point, they need a little miracle to turn things around in the mobile space.

 

But Intel and ARM SOCs are interchangeable, so these manufacturers can always go for the cheapest components, unless they put out Windows devices, instead of Android devices.

 

That has to be in other countries than the usa because i rarely see anyone without an ipad or a samsung tablet here in the states.

Whats really interesting is thet surface line up . I have the original pro and my gf had the original pro but now has the 3. People love the 3 and want to buy it but its to expensive for them.

I would think intel and ms can gain alot of market share if they have a top to botom surface line up

$400-600 gets you atom performance and $700 to $1,800 gets you core i5+ performance. They would gain alot of market share that way.

 

There are no-name tablets under $99 in the US too.

But a lot of them are being sold in emerging markets. You don't need a lot of power to surf, play some simple games, watch videos (esp. pirated videos). They don't need the latest OS either, if they're even running Android.

So the global tablet volume is supposedly being driven by these cheaper tablets whose brand names we don't see too much in the US. iPad share has been dropping because it's being overwhelmed by these cheaper devices. Apple lowered the price on some older models so the cheapest iPad starts at $249 for a 2-year old Mini. Not sure it'll make that much difference though, because that's still way more than $99 or even less.

 

Atom performance is going to start appearing more and more in the 100-200 USD range. With more upscale variants in the 200-500 USD range. I'm not sure Atom variants higher than 500 USD will exist for much longer. All the more interesting when you consider their company wide margins are still stratospherically high.

Fanless Broadwell is set to take the stage with tablets starting from 500 USD and possibly lower, I believe. It's somewhat amazing how fast Intel is trasitioning into the lower price brackets with their tablet chips. 2 years ago it was difficult to find an Atom tablet for less than 800 USD, with most of them being 900+ USD. A core based tablet usually started around 1500 USD and up back then.

With fanless Broadwell getting into Atom's power envelope, it may be that Intel will, as you noted before, keep performance the same but continue to shrink the chip (cheaper to produce) and lower its power consumption.

Broadwell (and followups) then take over the high performance tablet category to compete with Apple and Samsung. While the cost to manufacture Atom based chips for the low end continues to decrease while at the same time offering better and better battery life.

When Intel spoke of breaking even with their Mobile Chip division it may be that they expect the cost to manufacture Atom based chips will drastically decline with upcoming chips while boosting demand for higher margin chips.

There's also the possibility that if fanless Broadwell (or its followups) based tablets end up offering greater performance and lower power consumption than competing iPad solutions, that Apple could be forced into using Intel chips as they were when they were forced to transition to Intel CPUs for their Mac products.

A whole of IFs, and it isn't as if ARM is going to stand still while all of this is happening. It's just a question of whether Intel's progress will start to slow now that they are finally really serious about the segment, or if ARM can increase its progress to stave off Intel. Unfortunately, Arm (and their partners) doesn't have access to Intel's fabs, so they have to do better than just match Intel in design as they have to make up for the fab advantage that Intel has as well.

And there's still one thing that could possibly trip them up. Graphics performance. It is getting better certainly, but it's an area where the competition still has a good chance to one up them. Although with Nvidia doing all it can to slow the competition (in the mobile space) there, that could just play right into Intel's hands.

Regards,
SB

 

I'm not sure Windows tablets are the best way to illustrate Intel prowess and progress on the SOC front. I do prefer seeing Intel chips in Android or Chrome OS devices, I hope that Lollipop nullifies what seems to be a software disadvantage for Intel chips. Those tablets do not have the specs for run desktop apps properly (looking at my browser memory usage...), most desktop apps are not usable on touch screen, especially a tiny one => If anything they could serve as a huge display of the weakness of MSFT offering now: the lack of WinRT (/whatever they are called) apps. It can't be a good thing for the SOC powering those devices.

Looking forward, I see no trouble, they move and adapt a bit slowly but that is what you get from being such a huge corporation. On the other hand they can produce chips in shit tons, they can go really low in price, if necessary, etc. they won against every of their competitor like that by the way. Times are different for sure but lots of Intel advantage are still relevant to say the least.

Wrt to SOC I hope Intel simply ignores some of trends on the market (like Apple did with the many cores madness) and do their "own things'. Those Atom cores are a good basis, the biggest change I expect are in the cache hierarchy and the uncore.

It sounds right, I expect a Core II duo => Nehalem type of change. I expect Intel to add an extra level of cache (as Apple does).
The L3 will also be leverage by their (in house) GPU. As Intel is to provide SOC based on both their line of CPU, some standardization of the "uncore" make sense.

For the cores themselves I expect only refinements, better SIMD unit, then the unmatched cache hierarchy and memory sub-system will make the thing "fly".

EDIT
Silent Buddha indeed Intel has better fab, but they also have better designs, usually their cache hierarchy, the memory subsystem, /what seems the toughest parts to get right (looking at AMD failure to compete) are unmatched. Iirc those first gen baytrail still beat everything in benchmark related to "memory". Those CPU should perform slightly better in X64 mode.

I'm sure they are that "big" compared to the perfs they achieved. Intel did not disclosed anything and I can't find estimate of the die size and transistor count for the first generation of Silvermont chips.
EDIT
Found 102mm2 for the first gen bay trail (vs 89mm2 for Apple A8 on comparable process).

 

Care to provide evidence for that?

That will heavily depend on what is measured. For instance many SPEC benchmarks are compiled in 32-bit mode to get higher scores.

Also look at this post comparing nbench 32- vs 64-bit on Z3770.

 

Some things are odd in this article:

My understanding is that *all* BT get contra revenue, including those for Windows tablets.
Also the contra revenue was here to help offset extra HW costs that are supposed to be solved at some point in the future. I don't think Intel will have this solved real soon, so if they indeed stop contra revenues I guess we can expect low cost OEM to drop Intel due to higher costs.

 

The first article said Intel shipped 40 million tablet parts. Any of those on any name brands that people in the West would recognize?

Or did they all go on some Chinese brands mostly sold in China?

They presumably want to sell products in the premium part of the market at some point, to get the kind of margins they're used to getting.

 

Here in the UK (and Ireland, I think), the Hudl 2 tablet released by Tesco (the biggest supermarket) is a Baytrail device.

Wouldn't surprise me if this was one of the biggest-selling tablets over the festive period over here.