The Intel Atom Thread

[Abwx]

also low power sandy/ivy bridge with 847 a 1007U

you can buy a mini itx 1007U (dual core ivy bridge at 1.5GHz, 6EUs 17w) for around $90, and it's probably still a good amount faster... if bay trail dual core Celeron is just replacing it for the same price, it's certainly not great.


anyway, do bay trail have HT support disabled for now, or no HT planned at all?

Comsumption is high , yet some people by there think
that the same chip with double the cores and as high
frequency will consume three time less...

Anyway , for easy comparison :

 

[blackened23]

Comsumption is high , yet some people by there think
that the same chip with double the cores and as high
frequency will consume threeb time less...

Yeah that 75W peak of the A4-4000 is pretty high. No disagreements here. And folks were saying the 5000 could fit in a tablet, guess not. Maybe that's what the OEMs knew that we didn't.

Keep moving them goalposts, son! Comparing the desktop part which is intended to use more power (celeron J1750) to the tablet part which isn't tested (Bay Trail-T).

 

[Abwx]

Yeah that 75W peak of the A4-4000 is pretty high. No disagreements here. And folks were saying the 5000 could fit in a tablet, guess not.

The A4 4000 is a 65W DT Trinity , i guess they used it to create
some confusion ,and it worked , as proved by your sayings...

I edited my post with the 5000 numbers..

 

[Khato]

Comsumption is high , yet some people by there think
that the same chip with double the cores and as high
frequency will consume three time less...

Really? Comparing desktop power consumption numbers using an 860 watt power supply against notebooks? I'll give you the benefit of the doubt and assume that such wasn't noticed. Considering that according to this review - http://www.techpowerup.com/reviews/Corsair/AX860i/6.html - the power supply used shows a draw of ~11.5W from the socket with zero load.

In addition to that, we can look at other reviews of the Celeron G1610 and A4-4000 to arrive at expected delta power consumption between idle and load of roughly 20W and 40W, respectively. Comparing against the chart, we see a difference between the maximum and minimum recorded power consumption of roughly 10W and 35W - yet another very clear indication that the minimum consumption is close to idle and there is quite a bit of base power consumption present not directly attributable to the platforms under test.

 

[Abwx]

Really? Comparing desktop power consumption numbers using an 860 watt power supply against notebooks? I'll give you the benefit of the doubt and assume that such wasn't noticed. Considering that according to this review - http://www.techpowerup.com/reviews/Corsair/AX860i/6.html - the power supply used shows a draw of ~11.5W from the socket with zero load.

From its efficency curve we can extrapolate 75% efficency
at most at low powers , actualy once the PSU is loaded
parts of thoses 11.5W will be no more losts , so the plateform
absorb about 15W , but still , the Kabini 5000 above was tested
with a laptop , hence including the screen , thus , overall ,
theses two chips are on the same league TDP wise.

 

[blackened23]

From its efficency curve we can extrapolate 75% efficency
at most at low powers

So "extrapolation" is possible between different desktop, mobile, and tablet parts. Interesting. And here I didn't think you had a sense of humor. You're a funny dude!

 

[Abwx]

So "extrapolation" is possible between different desktop, mobile, and tablet parts. Interesting. And here I didn't think you had a sense of humor. You're a funny dude!

Far more possible than using Richland as basis to estimate
Kabini s tablets capabilities....

 

[Khato]

From its efficency curve we can extrapolate 75% efficency
at most at low powers , actualy once the PSU is loaded
parts of thoses 11.5W will be no more losts , so the plateform
absorb about 15W , but still , the Kabini 5000 above was tested
with a laptop , hence including the screen , thus , overall ,
theses two chips are on the same league TDP wise.

Why extrapolate in that method? We know that at zero load it uses 11.54W and when delivering 39.71W DC in their low load test it's consuming 51.39W - a delta of 11.68W. So we know that its efficiency at low power is going to be abysmal, as is typical for power supplies of this design. I'd be surprised if it was even hitting 50% efficiency for the J1850 board.

Anyway, I wouldn't even try to guess at numbers with the poor quality of the power information available in this review. It's barely even useful for comparison between the parts included in the graph due to the lack of a measured idle power. The only potentially interesting point in the J1850 power date is the ~10W spike, but without context there's not a lot to be said about it. (For example, at least half of that could be due to the combination of SSD and memory peak power consumption.)

 

[Shivansps]

Far more possible than using Richland as basis to estimate
Kabini s tablets capabilities....

Actually you cant do either, the Desktop BT-D aren't the same die as tablet BT-T, also dual core BT-D are actually quad core BT-D with 2 cores disabled, what may lead intro some power issues too.

BTW, J1750 igp runs at 750 not 792mhz.

 

[monstercameron]

Yeah that 75W peak of the A4-4000 is pretty high. No disagreements here. And folks were saying the 5000 could fit in a tablet, guess not. Maybe that's what the OEMs knew that we didn't.

Keep moving them goalposts, son! Comparing the desktop part which is intended to use more power (celeron J1750) to the tablet part which isn't tested (Bay Trail-T).

if an ivy i5 can then so can kabini

 

[Abwx]

Some people are expecting BT T to have double the core count
of this celeron and half the TDP despite similar frequencies ,
a hard landing in perspective , indeed...

 

[Khato]

Some people are expecting BT T to have double the core count
of this celeron and half the TDP despite similar frequencies ,
a hard landing in perspective , indeed...

Uh-huh. Just like some people are expecting the Pentium J2850 to have double the core count of this celeron and the same TDP... because, ya know, that's what the leaks have stated.

Oh, or are you again trying to claim that you can derive SoC power consumption from those graphs? And that Baytrail-T won't be able to have double the core count at half the power consumption of that fictitious figure? In which case I'll heartily agree for obvious reasons.

 

[blackened23]

Far more possible than using Richland as basis to estimate
Kabini s tablets capabilities....

So extrapolation is good when it suits your argument but when it doesn't, throw it out the window? Am I following you correctly? You CANT extrapolate performance or perf/watt among different architectures with different intended market segments. They are designed with completely different rules and constraints. To even suggest it is somewhat silly - it's like saying you can extrapolate core i5 ULV performance and efficiency from the real desktop core i5. We obviously know that isn't case since the core i5 ULV performs less but with far better efficiency.

 

[Abwx]

Uh-huh. Just like some people are expecting the Pentium J2850 to have double the core count of this celeron and the same TDP... because, ya know, that's what the leaks have stated.

Oh, or are you again trying to claim that you can derive SoC power consumption from those graphs? And that Baytrail-T won't be able to have double the core count at half the power consumption of that fictitious figure? In which case I'll heartily agree for obvious reasons.

At base frequency it s quite feasible but once the turbos
kicks in the dualies will have more headroom to clock at max
for a longer time.

Since the whole tablet using a 3770 consume 3-4W
just to do playback video it would be unreasonable
to expect the thing to consume only 5W at the SoC
level at max frequency with four fully loaded cores
and eventualy GPU.

Intel s reluctance to communicate about TDPs say it all,
actualy thoses tablet chips are 7.5-10W TDP for the most
performing siblings but since the used power in usual tasks
is far below theses levels they decided to focus on what
they deem usability for portable gear , that is , a few watts
power usage , that is certainly possible for BT but also for
the concurrence , and not only AMD...

 

[Khato]

At base frequency it s quite feasible but once the turbos
kicks in the dualies will have more headroom to clock at max
for a longer time.

For which product? If still talking about Baytrail-D then the Celeron J1750 and Pentium J2850 are both 2.4 GHz base parts with a 10W TDP, but one is dual core and the other is quad. Doesn't that say something about how representative TDP is of actual power conusmption?

Since the whole tablet using a 3770 consume 3-4W
just to do playback video it would be unreasonable
to expect the thing to consume only 5W at the SoC
level at max frequency with four fully loaded cores
and eventualy GPU.

Well, I would link hot hardware's preview - http://hothardware.com/Reviews/Betting-On-Bay-Trail-Intels-Atom-Overhaul-Tested/?page=5 - as evidence that a Baytrail tablet only uses 2.67W during 1080p video playback... But I already know what the response to that will be, in which case I must query - where did you come up with the 3-4W number? If throwing out the Baytrail-T preview power measurements then I'm not aware of any others?

Intel s reluctance to communicate about TDPs say it all,
actualy thoses tablet chips are 7.5-10W TDP for the most
performing siblings but since the used power in usual tasks
is far below theses levels they decided to focus on what
they deem usability for portable gear , that is , a few watts
power usage , that is certainly possible for BT but also for
the concurrence , and not only AMD...

That must be the reason why none of the ARM SoC companies communicate TDPs too, right? Or is Intel to continue being held to a different standard?

 

[Enigmoid]

At base frequency it s quite feasible but once the turbos
kicks in the dualies will have more headroom to clock at max
for a longer time.

Since the whole tablet using a 3770 consume 3-4W
just to do playback video it would be unreasonable
to expect the thing to consume only 5W at the SoC
level at max frequency with four fully loaded cores
and eventualy GPU.

Intel s reluctance to communicate about TDPs say it all,
actualy thoses tablet chips are 7.5-10W TDP for the most
performing siblings but since the used power in usual tasks
is far below theses levels they decided to focus on what
they deem usability for portable gear , that is , a few watts
power usage , that is certainly possible for BT but also for
the concurrence , and not only AMD...

The whole 1600p tablet used 3-4 w during playback at an unknown brightness setting.

Fixed.

 

[Shivansps]

What part of "is not the same SOC" do you dont understand? you cant directly compare power figures on BT-D and BT-T, is not the same chip, die sizes are not the same.
Also BT-T cant be 10W TDP or it could NEVER be in a fanless tablet.

Kabini and Temash on other hand, are the exact same thing but with diferent frecuencies.

 

[liahos1]

Samsung has no problem supplying the "app processor" to use your terms it just has no incentive to do so. Samsung has a similar processor to Qualcomm s600/s800 with their exynos gen 5 (often called exynos octocore).

The use Qualcomm s600 s800 for these parts have an integrated LTE modem or are bundled with the LTE modem. Qualcomm since they are effectively the only current supplier of LTE modems means Samsung must have a relationship with Qualcomm. If the s600 is bundled with a world class CPU that is sold cheaper than buying the exynos 5 then Samsung is going to use that CPU.

Samsung has many divisions of their company, if they can get a part cheaper outside the company istead of inside they are required to do so for that increases that divisions profitability and financials. Samsung mobile is separate from Samsung foundry.

In 3g markets you see Samsung use exynos 5. We will probably also see less Samsung using Qualcomm parts in the next 2 years with broadcom, Intel, and nvidia entering the LTE market.

you must have a hard time grasping fundamental economics.

Assume samsung was generating 90% yield. Do you think it would be cheaper for them to supply their own apps processor to samsung mobile and just buy the lte baseband from qualcomm or does qualcomm somehow have the economics to justify bundling an apps processor at a price below the cost for samsung to make it themselves AFTER paying TSMC a fab margin and AFTER keeping margin for themselves? bust out a calculator and think about it!

 

[Abwx]

For which product? If still talking about Baytrail-D then the Celeron J1750 and Pentium J2850 are both 2.4 GHz base parts with a 10W TDP, but one is dual core and the other is quad. Doesn't that say something about how representative TDP is of actual power conusmption?

Apparently they ll stick to 7-7.5W for higher performing tablets parts
down from the DT and NB 10W thanks to a few fused off functions
and surely less agressive turbos , memory and GPU speed and so on.

Well, I would link hot hardware's preview - http://hothardware.com/Reviews/Betting-On-Bay-Trail-Intels-Atom-Overhaul-Tested/?page=5 - as evidence that a Baytrail tablet only uses 2.67W during 1080p video playback... But I already know what the response to that will be, in which case I must query - where did you come up with the 3-4W number? If throwing out the Baytrail-T preview power measurements then I'm not aware of any others?

The 3-4W number , for the whole tablet i assume , comes from TR.

What we saw was very similar power consumption from one generation to the next. Both tablets tended to idle at about 2 W of power draw, and both used 3-4 W during video playback. Total system power draw is probably a bit higher during CPU- and GPU-intensive workloads, but we didn't get any full-platform power use numbers for such scenarios.

http://techreport.com/review/25329/intel-atom-z3000-bay-trail-soc-revealed/4

This say that at 2W iddling we have the screen comsumption
included , hence the delta is absorbed by the CPU/GPU only.

That must be the reason why none of the ARM SoC companies communicate TDPs too, right? Or is Intel to continue being held to a different standard?

Whatever each one ethic level we are not supposed
to be gullible with either.

 

[Imouto]

@liahos1 You're assuming that Qualcomm would sell its LTE modem alone at a reasonable price. Make it expensive and make your SoC more attractive.

Fundamental economics.

Qualcomm is holding a 97% LTE market share BTW.

 

[escrow4]

The bottom line is that no one cares about AMD, Intel, Nvidia or Microsoft in the mobile space.

Windows is currently holding about a 5% tablet market share. Bay Trail is targeting that market and that's beyond lame.

Only uninformed sheep (i.e standard consumers) still buy shiny shiny fruit products or Android, not realizing that Windows tablets give them vastly more versatility. And of course, sheep want toaster appliances. Now that ARM has woken Intel up, things will be vastly different within a year or two.

 

[Khato]

Apparently they ll stick to 7-7.5W for higher performing tablets parts
down from the DT and NB 10W thanks to a few fused off functions
and surely less agressive turbos , memory and GPU speed and so on.

The 3-4W number , for the whole tablet i assume , comes from TR.

http://techreport.com/review/25329/intel-atom-z3000-bay-trail-soc-revealed/4

This say that at 2W iddling we have the screen comsumption
included , hence the delta is absorbed by the CPU/GPU only.

But that preview, it's just providing the power figures according to the instrumentation that Intel setup for reviewers to use. How can that possibly be trusted? [/sarcasm]

Really though, what exactly makes the techreport measurements acceptable for reference while the others, e.g. 2.5 watts for the SoC while running multi-threaded Cinebench, aren't? I don't understand how we have have the reviewers that spent time with these systems saying that the SoC never went above 2.6 watts, but you're still talking roughly three times that for the performance that reviewers observed? Or are you saying that Baytrail can perform even better than the previews if it uses more power?

 

[Imouto]

Only uninformed sheep (i.e standard consumers) still buy shiny shiny fruit products or Android, not realizing that Windows tablets give them vastly more versatility. And of course, sheep want toaster appliances. Now that ARM has woken Intel up, things will be vastly different within a year or two.

Yet another typical tech forum answer.

Go back to 2008, please.

 

[krumme]

you must have a hard time grasping fundamental economics.

Assume samsung was generating 90% yield. Do you think it would be cheaper for them to supply their own apps processor to samsung mobile and just buy the lte baseband from qualcomm or does qualcomm somehow have the economics to justify bundling an apps processor at a price below the cost for samsung to make it themselves AFTER paying TSMC a fab margin and AFTER keeping margin for themselves? bust out a calculator and think about it!

I think you miss the purpose for samsung at this stage.

Running a highly profitable mobile business independency of suppliers and the ability to put pressure on qc is of high strategic and economic importance. More than some minor profit/loss on soc.
The a15 big little does just that - independancy and pressure.
Besides. We dont know the marginal cost for ss. The alternatives and so on.
Its quite more complicated than you present it.

Regarding cost i would assume tsmc was in fact far cheaper because of their superior business model in this market. But as mentioned there could be far more in the game.

They all want to be the big monopoly guy. And do all to prevent their own supplier to be it.

They all praise the free market because they want to use it. But does whatever in their power to prevent that market from working when they are the ones in power.

They praise modesty and low wages, unless its for the shareholders and top management that is.

 

[Abwx]

But that preview, it's just providing the power figures according to the instrumentation that Intel setup for reviewers to use. How can that possibly be trusted? [/sarcasm]

Really though, what exactly makes the techreport measurements acceptable for reference while the others, e.g. 2.5 watts for the SoC while running multi-threaded Cinebench, aren't? I don't understand how we have have the reviewers that spent time with these systems saying that the SoC never went above 2.6 watts, but you're still talking roughly three times that for the performance that reviewers observed? Or are you saying that Baytrail can perform even better than the previews if it uses more power?

TR says that a single core can pump 1W according
to their JS number , wich is a single threaded task ,
hence 4 cores can use 4W on peaks , far from
Anand s number wich should be of equal amplitude
given that CB is FP heavy and FP ops are power hungry
tasks.