Slashdot Mirror
Intel Claims Haswell Architecture Offers 50% Longer Battery Life vs. Ivy Bridge
MojoKid writes "As with any major CPU microarchitecture launch, one can expect the usual 10~15% performance gains, but Intel apparently has put its efficiency focus into overdrive. Haswell should provide 2x the graphics performance, and it's designed to be as power efficient as possible. In addition, the company has further gone on to state that Haswell should enable a 50% battery-life increase over last year's Ivy Bridge. There are a couple of reasons why Haswell is so energy-efficient versus the previous generation, but the major reason is moving the CPU voltage regulator off of the motherboard and into the CPU package, creating a Fully Integrated Voltage Regulator, or FIVR. This is a far more efficient design and with the use of 'enhanced' tri-gate transistors, current leakage has been reduced by about 2x — 3x versus Ivy Bridge."
That's fantastic. I love seeing efficiency, but I imagine that the screen would eat most of the battery life in consumer applications.
The biggest battery drain on my phone is always the display, followed by "Cell standby". How is a CPU and chipset able to promise a 50% increase in battery life when it's not even the biggest power user in the phone?
Is this seriously 50% increase in battery life? Or just 50% reduction in power usage by CPU? The article wasn't clear on this. I'm assuming the power usage thing.
So what? Amtrak has seat-side power for your computers. What you say does not make sense. (
Most Respectfully Yours Mark Allyn Bellingham, Washington
Amdahl's law applies to performance, not really geared toward power. Unless you would like to enlighten the class with a mathematical explanation of your assertion?
I browse on +1 so AC's need not respond, I won't see it.
If you're watching 3-hours of porn in a single sitting you're doing it wrong.
"People who think they know everything are very annoying to those of us who do."-Mark Twain
Yes, but they don't have a ground plug, so jacking off in the can while plugged in greatly increases the risks of severe electric shock.
The world's burning. Moped Jesus spotted on I50. Details at 11.
Without checking the source, i bet it is only the cpu/gpu/power thtat is getting lower values. It is the old intel story again. First it was the atom cpu that was supposed to be super low power. However they forgot to mention you needed a chitset along with it for the video networking pci that was not so super savy with power.
Now the cpu/gpu is super power savery. But the wifi/display/battery/2g/3g/nfc/audio/cam/gps might still drain your battery in 3 seconds.....
FIVR in the mornin' FIVR in the evenin', FIVR all through the night!
Facts are useless, they can be used to prove anything.
Is this a laptop only chipset, or does intel have goodies for those who like to be chained to their desks?
Or, very right.
Like most CPU's these days, they produce a lot of variants.
For this article they are likely talking about the "U" variant with 15W TDP.
http://en.wikipedia.org/wiki/Haswell_(microarchitecture)#Mobile_processors
You can't really compare that with the (or say in same breath) desktop "K" variant with 84W TDP (also has twice the cores and threads).
http://en.wikipedia.org/wiki/Haswell_(microarchitecture)#Desktop_processors
I am pretty sure the benchmarks will be wildly different. Anyway the summary makes it sound like it is all one thing. I am sure it will be very good and all, but I know I won't be getting one of those power saving versions. POWER! (To quote Clarkson)
Still waiting on that math.
I browse on +1 so AC's need not respond, I won't see it.
Maybe with the old fluorescent backlights, but not these days. A typical LED backlight on a laptop draws something like 3 watts at maximum brightness. It isn't lost in the noise, but it is by no means the main power draw. The CPU, chipset, and RAM take way more current.
Check out my sci-fi/humor trilogy at PatriotsBooks.
Math tip: A 50% increase in battery life (what they actually claimed) isn't the same as doubling it.
Also, since a big selling point for Haswell (aside from power efficiencies) is the claimed greatly improved (~2x for laptop-oriented models, ~3x for desktop-oriented models) improvement in graphics performance, I'd be very surprised if their claims for about battery life were focussed on systems using discrete GPUs rather than relying on the integrated graphics on Haswell.
Well, except that they explicitly claimed that was overall battery life, and it was a 50% increase not 2x, and they actually cited numbers for improvement in idle life and it was much higher than the +50% claimed overall (or even the 2x you pulled out of who-knows-where), since their claimed idle-mode improvement was twenty times (TFA is less clear on this, but Computerworld covers the same event with more specificity: "And in idle or standby mode the chips will do even better, extending battery life by up to 20 times, [Rani Borkar, Intel's Architecture Group VP] said." [emphasis added])
The analogy is sound, the "parallel" part is the processor and the "non-parallel" part the rest and it'll approach the same power baseline with increased processor efficiency as it does the performance baseline with increased parallelization. But I feel it's a rather silly complication of the obvious, unlike parallelization. Yes of course if the screen is the biggest power hog, then it has the most potential for improvement. Note that it would be a fallacy to think it will always have the greatest improvement, if the screen takes 60% and the processor 40% but you can only reduce screen power by 10% to 54% and the processor by 50% to 20% of the original you gain more with the processor.
Live today, because you never know what tomorrow brings
My point being is that Amadhl's law is not some generalized law of diminishing returns (we already nave one of those) and it's not 1-(1/) * 100, which the OP seems to think. It applies to a vary narrow set of problems in parallel processing. Can't just go around smearing it all over everything like a Canadian with a jar of mayo.
More to the point, however, the OP's assumption that the screen uses the most power is dead wrong.
I browse on +1 so AC's need not respond, I won't see it.
Source? This is the first I've heard of this, I haven't seen any articles on the subject, so this would be very enlightening. Generally Thinkpad quality is very high, even if their screen quality went to garbage starting around Thanksgiving 2012... It would be interesting to see more details on this, as I have been tracking the downward spiral of Thinkpad quality ever since the Lenovo CEO Yang Yuanqing announced that they were going to square off the Thinkpad vs Ideapad brands under lenovo at the cost of giving users worse quality products under both brands....
moox. for a new generation.
ugh. nave=have and that should read 1-(1/{insert some part specific performance factor here}) * 100
I browse on +1 so AC's need not respond, I won't see it.
here is an imaginary i5-3439Y laptop power budget: :1 W
screen : 12W
board
cpu : 15 W
HDD : 1W
Wifi : 1W
Total : 30W
To slash that by 50% you would have to have a magical CPU that consume no power, so there is your upper bound on power reduction...
That's nice and all, but that is not Amdahl's law.
I browse on +1 so AC's need not respond, I won't see it.
I wonder how the performance vs power consumption compares to the old Transmeta chips that started the trend.
Amdahl's laws are many.
Here are four of them.
0. Amdahl’s parallelism law: If a computation has a serial component S and a parallel component P, then the maximum speedup is (S+P)/S.
1. Amdahl’s balanced system law: A system needs a bit of IO per second for each instruction per second: about 8 MIPS per MBps.
2. Amdahl’s memory law: alpha=1: that is, in a balanced system the MB/MIPS ratio, called alpha, is 1.
3. Amdahl’s IO law: Programs do one IO per 50,000 instructions.
Corollary:
In any discussion of computer architecture, at least one member of the set of Amdahl's laws is bound to be relevant.
Really? Amdahl's law is:
Tn = a + (1-a)/N
Where Tn = Time with N cores
N = Number of Cores
a (should be alpha) = fraction of instructions in serial code.
What you are talking about is:
Bp = (1-((Pt - Pc)/pt))*100
While Amdahl is significant to the computer science world, are you claiming he invented percentages?
I browse on +1 so AC's need not respond, I won't see it.
Please provide the law that is relevant to this discussion of computer architecture.
I browse on +1 so AC's need not respond, I won't see it.
Soon motherboards will be just wiring for the I/O and CPU
The Zeitgeist Movement
Early last year some Lenovo Thinkpads had issues with lockups due to a voltage regulator being off spec.
Not terribly on-topic, but it was either that or scream: "I just bought an Ivy Brigde laptop dammit, Dammit, DAAAMMMIT!!!"
But putting the voltage reg in the CPU seems to be fraught with peril as well.
This means you are going to have to 1) have redundant regulation on the mo-bo for other components, and 2) subject your CPU to much higher (and unregulated) voltages. You've added another heat generation source right there on the CPU, and power excursions are likely to take out your processor.
Sig Battery depleted. Reverting to safe mode.
http://i01.i.aliimg.com/img/pb/419/444/420/420444419_371.jpg
"The Adobe Updater must update itself before it can check for updates. Would you like to update the Adobe Updater now?"
you forgot the 2-4w for RAM, if not more.
But higher voltages means less current, which helps.
Plus if the voltage regulators are in the CPU package, they can use the MUCH better thermal solution provided for it.
HI O WISE PRINCE. WHT TOOK U SO DAM LONG?
I've highlighted the problem with this refutation.
Check out the Lenovo forums regarding the "stop code" problem on the T430s model. They rectified the production problems in early September.
Incidentally, coming from Macbooks I have to say that press coverage of Windows/Linux systems and their performance issues is very scanty. It feels like no single model sells enough units to garner a critical mass of attention. With Apple stuff, every model has 3rd party teardown videos, other online guides and press attention just days after hitting the shelves. Maybe the difference is the typical Apple user cares more... I can't figure it out but wouldn't be surprised if the non-Apple segment was suffering from fragmentation.
Thinkpad quality is still some of the best in the business, and I think their low extended warranty prices are proof of that. But it does appear that the ideapad consumer focus has pulled them off track a bit; The 'Thinkpad' brand is being spun off into a separate division to address that problem.
Yes, but they don't have a ground plug, so jacking off in the can while plugged in greatly increases the risks of severe electric shock.
You've obviously never tried electric stimulation...
Ezekiel 23:20
If that's true then maybe Intel is making this move so they can sell more product: Power breakdowns to stand in as a replacement for technological obsolescence (which has been petering out in recent years).
And before anyone calls me cynical, I know for a fact that Intel is concerned about keeping the replacement cycle going. They have stated it at times when investors were getting jittery, and they even had a TV ad in plain view that admitted they wanted to entice people who "thought" they were perfectly happy with their existing PCs.
But only what you listed as the 0th law is credible and broadly accepted, as a fundamental rule. There is great doubt as to under what conditions those other 'laws' can or will continue to hold.
Is that marketing speak for "we were unable to increase the operating frequency"?
thegodmovie.com - watch it
Too bad CPU power consumption hasn't been the biggest consumer of watts in many years.
Hint; the biggest amount of consumed current in most laptops is the glowing part you look at.
The preceding comment is my own, and in no way construes an opinon of the Emperor of Mankind.
Don't hold your breath. He obviously could have just said that the screen eats enough power that you couldn't possibly cut total power consumption by half with just the CPU. But that doesn't make him sound as intelligent and mysterious as citing a mathematical argument for which he has no idea how it would actually work out without actual numbers and isn't really relevant in the first place.
laptops are dc and I don't think ground pass though the power brick to laptop.
The Y capacitor can leak enough for an uncomfortable tingle on sensitive skin like your bare lap (eg wearing shorts) or the underside of your forearms.
This post contains no rudeness or derision of any kind. All arguments are friendly. Terms and exclusions may apply.
Wait, I thought Adama's law was: The only good toaster is a dead toaster.
You already have a separate, programmable regulator for Vcore (overclockers fiddle with it all the time) and in both cases if the regulator fails the CPU is toast so there's no advantage in keeping it outside. I'm not sure how they integrated the reactive components, but they're surely more reliable than current electrolytics, plus shorter paths mean less voltage drop meaning less stress.
This post contains no rudeness or derision of any kind. All arguments are friendly. Terms and exclusions may apply.
This means you are going to have to 1) have redundant regulation on the mo-bo for other components,
Nope. Motherboards already had dedicated regulators just for the CPU.
High-speed CPU core logic needs very low supply voltages, around 1.0V these days. Lower speed parts built in older processes need higher voltages -- 1.2V, 1.5V, 1.8V, or more. There's not much on the motherboard which even can share supplies with the CPU. Also, CPUs now dynamically vary their own core voltage (by sending commands to the regulator) in order to save power. That wouldn't work so well with other chips sharing the same regulator.
It's been a very long time since the regulators which powered the CPU core also powered any other chips on the motherboard.
and 2) subject your CPU to much higher (and unregulated) voltages. You've added another heat generation source right there on the CPU, and power excursions are likely to take out your processor.
Who said the input to a regulator must be unregulated? Current x86 systems already convert one regulated voltage to others. The ATX power supply outputs regulated 12V DC to the motherboard, which is then converted to several low voltages to supply the CPU and other devices. Haswell's integrated voltage regulators accept 1.8V input, so in practice they're going to be fed by a 1.8V regulator on the motherboard.
Heat doesn't appear to be that big a deal. The top desktop TDP bin is going from 77W (Ivy Bridge) to 84W (Haswell). The total system power is going down, not up -- despite the extra conversion step, the integrated voltage regulators offer efficiency gains.
(One is that it's now practical to have a lot of independent power domains. Haswell apparently has no less than five independent power planes. This makes it possible to dynamically adjust the voltage of different portions of the chip independently, giving a better match between activity level and power use. Another gain is that it's difficult and inefficient to supply a very low voltage at very high current over long distances. This is why nobody was ever even slightly interested in extending the ATX spec to supply ~1.0V directly to the CPU. Haswell puts the high current / low voltage supply as close to the load as it possibly can go.)
AMD CPUs run 4.4GHz stock. There must be a different reason. It might be a tradeoff between complexity and pipeline depth.
thegodmovie.com - watch it
Considering that everything is ultimately running off a battery that provides only a single voltage, the distinction is moot.
Check out my sci-fi/humor trilogy at PatriotsBooks.
Or its simply the fact that dynamic power is related to the switching speed via the square of the voltage * frequency. So, increasing the clock and voltage causes the power to go up significantly for a given number of transistors.
So, its an evil tradeoff, add more transistors increase leakage, or bump the clock rate and increase the dynamic power.
AMD has chosen the faster clockrate, for a few percent decrease in efficiency, while Intel has chosen the save power option for a few percent decrease in performace. It makes sense as intel is top performance dog right now. Being top power dog is now their focus. If AMD starts getting close they will release another i7 EE with more memory channels and clocked another Ghz faster.
In both cases, I'm sure they could design a CPU with a TDP of 300W and give themselves a clockrate bump to match the 5.5Ghz zSeries IBM sells. Of course then they would need a huge blower like IBM uses. I'm sure that if either of them thought they could get away with selling a few $500k CPUs to fit in a $20 million machine they would do it.
I thought the Y-shaped capacitor was what allowed time travel.
For workloads where the system is awake but mostly idle (think web browsing etc) you'll see enormous gains in energy efficiency; the less idle it is the less gain.
When I browse the web on my Nexus 7 tablet, "Screen" already takes at least 67 percent of the battery. And that's with ARM, which already sips less power in general than x86. What CPU upgrade will fix that?
Then perhaps the next step is to build user interfaces that aren't based on scrolling or other smooth motion, so that something with a laptop form factor and an e-ink display becomes viable.
Now they can make the OS and application coding less efficient!
You make a good point about Haswell using less power at idle. And an x86 CPU that uses less power at idle could help Windows tablets escape the RT jail that Microsoft has erected around the ARM architecture. I was just trying to point out that "enormous gains in energy efficiency" for the CPU won't necessarily translate to "enormous gains in energy efficiency" for the whole system, especially as laptop makers shift their emphasis away from 10" subnotebooks to the larger screen models.