The Intel Atom Thread

[Abwx]

As long as turbo can be turned on/off by software, I don't see why it would be counter-productive.

Read again my post , it says that it s counter productive
when running on battery but usefull when main connected ,
wich is not the same as just saying that it s counter productive.

 

[NTMBK]

Read again my post , it says that it s counter productive
when running on battery but usefull when main connected ,
wich is not the same as just saying that it s counter productive.

Why would it be counter productive on battery? Turbo lets you keep the CPU at a lower clock most of the time, racing up to a fast clock briefly to let you quickly perform a task before returning to a low power state. It gives you responsiveness on demand without tanking the battery. It's a concept know as "Hurry Up and Go to Sleep", or "HUGS".

 

[Abwx]

Why would it be counter productive on battery? Turbo lets you keep the CPU at a lower clock most of the time, racing up to a fast clock briefly to let you quickly perform a task before returning to a low power state. It gives you responsiveness on demand without tanking the battery. It's a concept know as "Hurry Up and Go to Sleep", or "HUGS".

Not only it s not a free lunch but you ll have
to pay at prestige restaurant prices for some
mac donalds stuff.

To reduce the task length by 30% you ll have
to increase CPU comsumption by 80-100%.

 

[drikkie]

But with mobile you don't only use less cpu but can shutdown the screen that much more early.

 

[Abwx]

But with mobile you don't only use less cpu but can shutdown the screen that much more early.

Undoubtly.

 

[mikk]

You dont get official numbers because they would contradict
what was displayed to the press...

People here seems to not suspect the obvious , that is that
Intel binned a few chips out of dozen thousands ones to get
a few samples that could run at significantly lower voltage
than the average of their production.

Selecting chips that can work with about 20% lower voltage
will reduce the cores power comsumption by 44% , hence
if the chip is basicaly a 7W SoC you ll end with only 4.86W...

Intel didnt publish numbers because they know that the
delivered chips wont be as efficient as the golden samples
they used for their marketing reviews, they simply cant
garantee the specs they displayed with an unknown protocol.

Then it is even more strange that you are still waiting for TDP numbers when you know it won't happen. Also your "didnt publish numbers because" is another claim you cannot prove as usual. Maybe SDP makes more sense for this particular use case. There is no TDP industry standard. TDP doesn't necessarily mean max power, look to AMD.

 

[Nothingness]

TDP doesn't necessarily mean max power, look to AMD.

Or look at Intel Haswell breaking its TDP by almost 100% This behavior is even documented in Intel datasheets.

I'm getting mad at waiting for real BT devices and reviews, want to know what battery life really is and if Linux can be installed and if 64-bit is available...

 

[Khato]

Technicaly speaking a turbo feature is counter productive
on a mobile device running on battery but usefull when
connected to the main.

Technically correct. But at least with Intel's implementation you get a hardware based system that doesn't lag demand by a few hundred milliseconds like the competition. And Intel's turbo frequencies are still within a reasonable thermal budget while the competition happily raises voltage well beyond what it should just to get a fraction more performance.

Oh, and as everyone else has already said, the technicality is pretty much entirely invalidated for the simple fact that chips run at the ideal performance per power point on the curve would never sell because of how horribly slow they'd be by comparison. It's necessary to have DVFS of some manner in these products, regardless of what it's called.

early numbers we got from a few
manufacturers all point to the intel "reviews" as being
doctored

Mind providing a link to those early numbers from manufacturers? The only guidance I recall seeing from manufacturers was that they were pleased with both performance and power - http://www.digitimes.com/news/a20130807PD222.html

 

[liahos1]

Jacob Hugosson ?@JacobHugosson 13 Oct
Asus "up to 11 hour" claim with #TransformerBook #T100 seems to be correct during browsing #BayTrail

joebelfiore ?@joebelfiore 11 Oct
got a Surface Pro 2 to test (thx @panos_panay). Been on a plane for the last 5.5 hours, doing email THE WHOLE TIME. Battery at 42%. Yay!


the last one seems interesting. I dont know what version of haswell surface pro 2 but from what i understand its not the fanless version. doing some quick math

- 5.5 hours / (1-.42) = 9.5 hours battery life
42.5whr battery / 9.5 = 4.47w during joe belfiore's email athon

using MSFT's 8hr battery life target

- 42.5/8hr = 5.3w

seems really good for the performance it comes with. I wonder what kind of real world performance we get with fanless haswell and then broadwell.

using dell's / asus quoted battery life of 11 hours on baytrail 10.1 devices with a 30whr battery

30/11 = 2.7w. The performance per/watt of these low voltage haswells and baytrail look really interesting esp vs a15 vanilla stuff we've been seeing.

Also - doesn the ipad have a battery of the same capacity as the surface pro 2?
a
42whr battery / 2.7w= 15.6hrs of battery life. Will anyone make this ?

 

[blackened23]

Or look at Intel Haswell breaking its TDP by almost 100% This behavior is even documented in Intel datasheets.

I'm getting mad at waiting for real BT devices and reviews, want to know what battery life really is and if Linux can be installed and if 64-bit is available...

You can't "break" TDP. TDP is NOT a measure of maximum power consumption and never has been. It is, quite funny to see the SDP / TDP argument going on for pages upon pages upon pages, though, when there is no industry standard for the definition of "TDP" yet some here act like there is a standard.

Like I said earlier, the bottom line is that Bay Trail is making it into razor thin 7 inch tablets while Temash is not. Temash is a laptop chip unless you gimp it beyond belief in terms of Turbo and clockspeeds to the point of being worthless in a thin tablet, I suspect OEMs realized this. Not that there's anything wrong with a chip for laptops and convertibles - hey that's great, pretty good perf/price in a convertible or laptop. But at some point you just have to capitulate and realize that one product is better than another for a given (thin tablet) form factor.

 

[beginner99]

Like I said earlier, the bottom line is that Bay Trail is making it into razor thin 7 inch tablets while Temash is not. Temash is a laptop chip unless you gimp it beyond belief in terms of Turbo and clockspeeds to the point of being worthless in a thin tablet, I suspect OEMs realized this. Not that there's anything wrong with a chip for laptops and convertibles - hey that's great, pretty good perf/price in a convertible or laptop. But at some point you just have to capitulate and realize that one product is better than another for a given (thin tablet) form factor.

I don't disagree with you but maybe BT is just cheaper than temash...would not be surprising at all.

 

[CHADBOGA]

Why would it be counter productive on battery? Turbo lets you keep the CPU at a lower clock most of the time, racing up to a fast clock briefly to let you quickly perform a task before returning to a low power state. It gives you responsiveness on demand without tanking the battery. It's a concept know as "Hurry Up and Go to Sleep", or "HUGS".

When the Phenom II didn't have Turbo, according to the usual suspects, Turbo was a scam, a gimmick, a sign of a company's immorality(I kid you not).

When AMD implemented Turbo in its desktop processors, it became a non-issue altogether.

History repeats.

 

[SiliconWars]

Bottom line is that Bay Trail is going into super sleek 7-8" tablets, and Temash is powering really cheap, sub-par laptops.

Everything else is, quite frankly, academic.

Bottom line is AMD is getting paid more for each chip and selling millions more of them.

 

[mikk]

I don't disagree with you but maybe BT is just cheaper than temash...would not be surprising at all.

Too cheap prices? :biggrin: Funny given that not long ago people had doubt that Intels pricing policy could hinder a Bay Trail success. The reason is Temash isn't suited for tablets. Even if AMD would sell the A4-1200 for 20 USD less than Z3770 it's not appealing to OEMs.

 

[blackened23]

Bottom line is AMD is getting paid more for each chip and selling millions more of them.

Uhhhh? Wait, wait. You're saying AMD will sell more Temash chips than intel will sell Bay Trail chips. Are you a betting man?

 

[SiliconWars]

Too cheap prices? :biggrin: Funny given that not long ago people had doubt that Intels pricing policy could hinder a Bay Trail success. The reason is Temash isn't suited for tablets. Even if AMD would sell the A4-1200 for 20 USD less than Z3770 it's not appealing to OEMs.

The problem isn't the A4-1200, it's all the cheaper and competitive ARM chips.

Intel is going to have a real problem selling these new Atoms in any segment. They have no chance in phones - in fact they've already lost the K900 to Qualcomm - Lenovo Drops Intel for Qualcomm in Latest K-Series Phone

They can't beat AMD on price in craptops and graphics are much worse so the OEM's will be taking the AMD chips first anyway. Tablets look like Bay Trails best chance but the market is saturated and Bay Trail does not have any compelling advantages for the price.

I have no idea why this SoC is so large either. That would have been Intel's best chance to compete on price but at 110mm2 it's about the same size as any other.

Wall Street isn’t buying Intel’s mobile hype

 

[SiliconWars]

Uhhhh? Wait, wait. You're saying AMD will sell more Temash chips than intel will sell Bay Trail chips. Are you a betting man?

No I'm saying AMD will sell more Temash than Intel will sell Bay Trail tablet chips.

 

[liahos1]

The problem isn't the A4-1200, it's all the cheaper and competitive ARM chips.

Intel is going to have a real problem selling these new Atoms in any segment. They have no chance in phones - in fact they've already lost the K900 to Qualcomm - Lenovo Drops Intel for Qualcomm in Latest K-Series Phone

. They can't beat AMD on price in craptops and graphics are much worse so the OEM's will be taking the AMD chips first anyway. Tablets look like Bay Trails best chance but the market is saturated and Bay Trail does not have any compelling advantages for the price.

I have no idea why this SoC is so large either. That would have been Intel's best chance to compete on price but at 110mm2 it's about the same size as any other.

Wall Street isn’t buying Intel’s mobile hype

the k900 has been out for two quarters. lenovo and other oems put out multiple skus a year - so nothing to see here. Lenovo has worked with mediatek in the past along with intel. they'll probably put out a k920 sku with merrifield. along with other oems. remember nobody wants to be beholden to qualcomm and intel is the only real credible LTE competitor.

bay trail die size is 102mm^2 dont know where u got 110.

No compelling advantage for the price? How about it being faster and lower power than all the tier 1 soc's out there and being priced like its a mediatek chip? sounds pretty compelling to me.

since when were graphics the defining constraint with oems for SOC decision making? I honestly dont know how AMD has come up so often in this thread. the company is a turd, their products are turds. Remember all the hype with trinity? Bulldozer? all turds.

 

[Khato]

Intel is going to have a real problem selling these new Atoms in any segment.

I'd agree with that statement if they were actually charging the prices listed on ark for their Baytrail-T SKUs... but all indications are that they're going for around $10, not $32 to $37. (This was discussed a few pages back.) At that point, why pay $20 (or more) for a Snapdragon 800 or Tegra 4 when Baytrail is easily good enough and better in many respects?

As for smartphones, well, let's wait and see what happens with Merrifield? Since yeah, Intel's current phone offerings are still quite outdated.

 

[Vesku]

I thought OEMs were asking for the $10 price which indicated Intel was asking ~$15 for lowest model and somewhere between $20-$30 for the top end -Ts.

 

[SiliconWars]

bay trail die size is 102mm^2 dont know where u got 110.

Forgive me for being 8mm2 out.

No compelling advantage for the price? How about it being faster and lower power than all the tier 1 soc's out there and being priced like its a mediatek chip? sounds pretty compelling to me.

Meanwhile back in the real world, the only interest in it is from their usual x86 partners, who can't seem to get a decently specced tablet out for a decent price.

since when were graphics the defining constraint with oems for SOC decision making?

When performance and power is within 10% of each other yet one has graphics that are twice as good, what do you think OEM's would rather have?

I honestly dont know how AMD has come up so often in this thread. the company is a turd, their products are turds. Remember all the hype with trinity? Bulldozer? all turds.

AMD comes up so often because others a lot like you are obsessed with them. I strongly suspect it's due to AMD beating Intel in the past with far less resources, and how that doesn't fit your world view. Now you are you seeing the reality of the situation - that is Intel is not competitive vs companies their own size. Yes the chips themselves are reasonably competitive vs older 28nm ARM chips, but they can't compete on price. If you can't compete on price, you can't compete.

 

[SiliconWars]

I'd agree with that statement if they were actually charging the prices listed on ark for their Baytrail-T SKUs... but all indications are that they're going for around $10, not $32 to $37. (This was discussed a few pages back.) At that point, why pay $20 (or more) for a Snapdragon 800 or Tegra 4 when Baytrail is easily good enough and better in many respects?

If they are going for 10 bucks then great, they'll sell by the barrowload. There ain't no way Intel is even close to break-even with a 102mm2 die selling for 10 bucks though so that's a pyrrhic victory if ever there was one.

 

[liahos1]

I thought OEMs were asking for the $10 price which indicated Intel was asking ~$15 for lowest model and somewhere between $20-$30 for the top end -Ts.

from what i heard baytrail T is being priced aggressively (8-10 dollars).

I've heard this directly from analysts at Morgan Stanley (Joe Moore) , a friend of mine who recently came back from Taiwan and would be in the know after meeting with several oems and from what Renee James hinted at during a lunch with Patrick Wang at Evercore during IDF. Who knows what the truth really is with regard to $ price, but the general consensus appears to be the company is being aggressive here.

 

[liahos1]

If they are going for 10 bucks then great, they'll sell by the barrowload. There ain't no way Intel is even close to break-even with a 102mm2 die selling for 10 bucks though so that's a pyrrhic victory if ever there was one.

I dont think this is correct using very simple math. A typical 28nm wafer TSMC sells for $4.4k and generates a 48% GM on it. The cash gross margin is even higher given a fair bit of the COGS are depreciation and amortization. So for TSMC the wafer costs are $2.2k (embedding D&A). Assuming similar yield, intel's cost (including D&A) to produce a 102^mm2 die is on the order of magnitude of $4 dollars. the marginal cost of producing the chip should be much lower. Just saying.

 

[SiliconWars]

from what i heard baytrail T is being priced aggressively (8-10 dollars).

I've heard this directly from analysts at Morgan Stanley (Joe Moore) , a friend of mine who recently came back from Taiwan and would be in the know after meeting with several oems and from what Renee James hinted at during a lunch with Patrick Wang at Evercore during IDF. Who knows what the truth really is with regard to $ price, but the general consensus appears to be the company is being aggressive here.

A 102mm2 chip has ~586 die candidates per wafer. Even at 100% yield, at $9 a chip that means Intel is getting $5275 per wafer. Simply put, this pricing is not sustainable.