Intel has $55.9B record year, ships 46M tablets

[386DX]

My question would be; isnt the power draw figure for the whole platform, not just the chip? Why blame the consumption of the whole platform (screen, speakers, SoC, memory, etc.) on just the SoC?

Or am I wrong and they're measuring just the chip as it runs?

Notebook review and any other can not measure the actual power draw of the SoC directly. Even the entire platform power is hard to measure. The only accurate way of measuring the SoC power draw was probably the setup Intel had before where they had probes directly connected to the SoC.

This is why I laugh when people quote notebook reviews power draw numbers and conclude AMD chip uses less power because the watt meter from the wall shows less, yet the actual run time on battery is also less. Let's use the Asus transformer as an example they measured a 12.8W from the outlet the AC adapter was connected to... What conclusion can you draw about the SoC or platform power consumption? Really nothing because if I plug the exact same Asus transformer to a USB port on a computer and measure the power draw I'll get 2.5W. So does it draw 2.5W or 12.8W? Nobody really knows but it's some where in between for sure. At 2.5W you'll probably see the battery on the device slowly going down as the 2.5W isn't enough to charge the battery and power the device. At 12.8W you'll see the battery charge while you use the device. If it came with a 17W adapter you'll probably see the battery charge even faster, but the SoC itself isn't using any more power.

 

[Abwx]

At 12.8W you'll see the battery charge while you use the device.

The battery was charged at 100% before doing the tests...

The device draw 11.8W with a 10W adaptator, that say it all, power drain is limited by the adaptator, not the SoC.

 

[Khato]

The battery was charged at 100% before doing the tests...

The device draw 11.8W with a 10W adaptator, that say it all, power drain is limited by the adaptator, not the SoC.

How about we put an end to this subject with the notebookcheck review of the T200?

http://www.notebookcheck.net/Asus-Transformer-Book-T200TA-Convertible-Review.127091.0.html

Measured maximum load of 12.5/11.7 watts with a 33 watt power adapter and a faster Baytrail SKU.

 

[Abwx]

How about we put an end to this subject with the notebookcheck review of the T200?

http://www.notebookcheck.net/Asus-Transformer-Book-T200TA-Convertible-Review.127091.0.html

Measured maximum load of 12.5/11.7 watts with a 33 watt power adapter and a faster Baytrail SKU.

Yes, this one has a slightly better power management at the expense of performance, of course :

The clock fluctuated between 1.5 and 1.7 GHz during multi-core load. The Turbo range is apparently out of reach. However, it was a stable 2.26 GHz in single-core load. Thus, 1.5 to 1.7 GHz seems to be normal during multi-core calculations.

1.5-1.7 is about the perf of a 1.2-1.4 Beema/Mullins/Kabini, besides you can see that Intel s 4W TDP is largely exceeded by all bay trails.

 

[Khato]

Yes, this one has a slightly better power management at the expense of performance, of course :

The clock fluctuated between 1.5 and 1.7 GHz during multi-core load. The Turbo range is apparently out of reach. However, it was a stable 2.26 GHz in single-core load. Thus, 1.5 to 1.7 GHz seems to be normal during multi-core calculations.

Which would be why it performs better than the T100?

How much faster is the Atom Z3775 (1.46 GHz) in the T200TA compared with the Atom Z3740 (1.33 GHz) in the Transformer Book T100TA? Cinebench R10 recorded an 18% increase in the multi-core test and 15% in the single-core test. The GPU can even present a plus of 21% (Shading 32-bit).

1.5-1.7 is about the perf of a 1.2-1.4 Beema/Mullins/Kabini, besides you can see that Intel s 4W TDP is largely exceeded by all bay trails.

And this is why I typically don't even bother with you any more. First, with respect to performance, you know how AMD's reference platform compares to the T100, and how the T200 compares to the T100, so what's your point again? And as for the claim that ~12W maximum system load power at the socket implies that the SoC is using more than 4W... okay, believe what you will, the rest of us recognize that the remainder of the system could easily constitute 8W of power in that scenario.

 

[elemein]

Notebook review and any other can not measure the actual power draw of the SoC directly. Even the entire platform power is hard to measure. The only accurate way of measuring the SoC power draw was probably the setup Intel had before where they had probes directly connected to the SoC.

This is why I laugh when people quote notebook reviews power draw numbers and conclude AMD chip uses less power because the watt meter from the wall shows less, yet the actual run time on battery is also less. Let's use the Asus transformer as an example they measured a 12.8W from the outlet the AC adapter was connected to... What conclusion can you draw about the SoC or platform power consumption? Really nothing because if I plug the exact same Asus transformer to a USB port on a computer and measure the power draw I'll get 2.5W. So does it draw 2.5W or 12.8W? Nobody really knows but it's some where in between for sure. At 2.5W you'll probably see the battery on the device slowly going down as the 2.5W isn't enough to charge the battery and power the device. At 12.8W you'll see the battery charge while you use the device. If it came with a 17W adapter you'll probably see the battery charge even faster, but the SoC itself isn't using any more power.

Exactly. So why is everyone saying "Intel is lying, the SoC consumes over 10W, not even close to it's TDP" when its so obvious that NBC can only measure the entire platform's consumption?

 

[coercitiv]

So why is everyone saying "Intel is lying, the SoC consumes over 10W, not even close to it's TDP"

It's mostly a complete disaster information wise due to a combined number of factors:

• Intel used SDP instead of TDP for these chips.
• AMD used TDP but did not offer clear power consumption data.
• Reviewers used system idle/load power delta to approximate SoC power usage.
• Fans cherry picked reviews to paint their favorite brand in a better light. Keep in mind we are talking about both sides here.

 

[Nothingness]

And in the end who cares about TDP/SDP? What matters is how long your device lasts on its battery while doing what you want it to do.

 

[witeken]

Indeed, Intel is again punished for being so open.

Cf. Qualcomm, Apple, MediaTek, Nvidia, Samsung.

 

[sckmcck]

Shouldnt the first factors be 3.5 x 10^9..?.

Ooops, you are right! That's another three orders, so 150uW minimum to switch 1% of the transistors.

 

[witeken]

so 150uW minimum to switch 1% of the transistors.

Why only 1% of the transistors?

 

[elemein]

It's mostly a complete disaster information wise due to a combined number of factors:

• Intel used SDP instead of TDP for these chips.
• AMD used TDP but did not offer clear power consumption data.
• Reviewers used system idle/load power delta to approximate SoC power usage.
• Fans cherry picked reviews to paint their favorite brand in a better light. Keep in mind we are talking about both sides here.

So the best way to approach the matter would be to toss out pretty much all info we have from the manufacturer regarding TDP, SDP and power consumption and just measure the devices' power consumption and estimate the platforms consumption off that, no?

 

[Idontcare]

Why only 1% of the transistors?

Typical instruction pressure? How many circuits on a core are being utilized at the exact same time? Not many I'm guessing.

 

[III-V]

Typical instruction pressure? How many circuits on a core are being utilized at the exact same time? Not many I'm guessing.

Yeah, I remember listening to Nvidia's exascale project, Echelon or whatever it's called, and they mentioned that. If you light up all the transistors on a chip, you are going to melt a hole through your home's foundation real fast. Not literally, but yeah, not good for the power bill.

To build an ExaScale machine in a power budget of 20 MW requires a 200-fold improvement [from 2011] in energy per instruction: from 2 nJ to 10 pJ. Only 4x is expected from improved technology. The remaining 50x must come from improvements in architecture and circuits.

 

[coercitiv]

So the best way to approach the matter would be to toss out pretty much all info we have from the manufacturer regarding TDP, SDP and power consumption and just measure the devices' power consumption and estimate the platforms consumption off that, no?

The best way to approach the matter is to corroborate info from the manufacturer with measured review data and have a reasonable discussion.

What I did in my last post was to point out the trifecta that lead to the current situation:

• manufacturers gave little info on power consumption
• reviewers failed to measure power SoC power consumption in a meaningful way
• some forum members monopolized discussion around favorable examples

If you want my opinion, since I'm probably asked to choose a side here, I think Bay Trail is a nice chip but I have my doubts about the contra revenue program. One can't argue for the need of a contra revenue incentive in order to make the chip competitive price wise, and then end up pushing so much money that x86 tablets and phones end up cheaper than their ARM counterparts.

However, my doubts are just this, doubts. I trust other chip manufacturers to voice their concerns in public if that is indeed the case.