Intel Core Ultra 200V promises Arm battery life without compatibility issues
arstechnica.com
arstechnica.com
To me the main contributing factors to power efficiency are the architecture, very complex on x86, and the manufacturing process, which has hit a miniaturization wall.
All the low hanging fruits have long been taken on x86. How did they achieve this?
> To me the main contributing factors to power efficiency are the architecture
Also is this really true?
The real problem isn't actually the power-draw while working though. Where x86 processors run into real battery life trouble is that modern Windows and Linux are really bad about keeping power draw low when the laptop lid is closed.
You can put your phone to sleep and leave it for a couple days and find its only lost 10% battery life, but if you close a Linux or Windows laptop and leave it for a couple days without actually powering it off, you're typically looking at something more like 50% battery life lost.
This is the real problem that needs to be solved, and it's going to be quite difficult because it'll involve hunting down tonnes of super old crufty low-level OS code and redesigning it.
One of the biggest reason ARM laptops tend to not have this problem is that a ton of low-level OS stuff had to be totally rewritten for them, so they just did the efficient thing in the first place, rather than having to sift through existing, working code and tweak it for better efficiency.
If you optimize the software in such a way that the power draw on idle can be very low, all that work will go to waste if the cpu is clocked at 8w minimum. You need a cpu that supports going to that low wattage.
Taking this as a reference for an arm cpu on a laptop: https://www.anandtech.com/show/17024/apple-m1-max-performanc...
You can see that even with the laptop on showing a static screen, the power draw is less than 1W. That's crazy and tbh, I'm not sure I believe it myself.
In summary, I do agree with you that we need to optimize our OSs so that they become more efficient, but we also need the hardware to improve. 8w minimum is not nearly enough for good battery life.
Existing Intel Meteor Lake CPU in laptops already draw anywhere from 4W to 11W idle (That's CPU + screen in benchmarks).
Like these:
https://www.notebookcheck.net/Xiaomi-RedmiBook-Pro-16-2024-r...
That power draw has nothing to do with the CPUs power efficiency, though: Laptop lid closed usually means S3, though, and the CPU is completely unpowered in that state. (CPU registers are saved to RAM.)
The biggest power draw in S3 is the RAM itself, which has to be continuously refreshed to not lose its data, but that can be avoided by doing suspend-to-disk instead.
Modern systems usually do some hybrid between the two, e.g. suspend to both RAM and disk, and power off the RAM after a few hours (or at a given battery level) to meet a compromise between resume speed and battery life.
Flabbergasting, that this issue has not been addressed.
When I last used Linux and Windows on a Thinkpad, lid closed was equivalent to S3, as indicated by the little crescent moon LED on the lid, fans out, and the battery in that state would last for several days.
My Mac does occasionally turn on with the lid closed to do "useful background tasks" (not actually that useful to me, except for broadcasting "Find my Mac" via Bluetooth), but except for one time when I accidentally had that option permanently flagged one via some "prevent sleep" app (I only found out when I noticed that I could SSH to my laptop while it was in my backpack!), it hasn't noticeably drained my battery.
There is a real difference, but it’s nothing intrinsic to ARM or x86 I agree.
The TDP has nothing to do with the battery lifetime in normal usage. Any smartphone or laptop will drain completely its battery in a couple of hours at most when running at the power limit (you will never see a laptop configured for an 8 W TDP also having a 99 Wh battery; such batteries are normally used for laptops where the CPU is configured for a 45 W TDP).
For the battery lifetime what matters most is the power consumption when the CPU is idle (which should be a fraction of a watt, so that the total idle power consumption is dominated by the display power consumption) and also how big is the additional power consumption caused by light CPU loads, like Internet browsing, movie watching, document reading and editing etc.
* Averaged over some time window relevant for cooling
reported TDP: 125W
actual power usage: 350w
In this case the CPU consumes as much it can before the temperature exceeds the temperature limit. The actual power that is consumed will be determined by how good is the used cooler.
The MB manufacturers do this so that their motherboards will win the benchmarks in published product reviews.
Unless the BIOS locks the power control registers, you can configure any Intel or AMD CPU for any power limit you desire and the power consumption will never exceed the set limit.
A low power limit will diminish the achievable multithreaded performance, but unless the limit is very low the multithreaded performance per watt will be greatly improved.
The TDP matters a lot for a laptop or for a SFF computer. In this case the laptop or small computer is really designed for a definite TDP, e.g. for 28 W or for 45 W, and attempting to make the CPU consume more than that will result in thermal throttling.
For a desktop, the design of a motherboard does not depend on the TDP, because it does not include the cooling system. The MB designer only takes care that the voltage regulator should provide enough current to reach the highest power consumptions that will not destroy the CPU.
The actual power consumption that can be reached without thermal throttling is determined by the case, the case fans and whatever CPU cooler is used. For most desktops it is easy to ensure that the CPU can consume more than its nominal TDP, without overheating.
Appolagies for the confusion.
As for where I got my 8W figure from:
> All of the chips except for the Core Ultra 9 default to 17 W base power, with a minimum of 8 W (if the manufacturer configures it that way) and a maximum Turbo Boost power of 37 W.
As an analogy a 200W-equivalent LED lamp uses like 5W.
TDP is about thermal envelope, which is a factor of the power input but not the useful output.
Perf/watt would be a better point for your analogy. But TDP doesn’t care about perf/watt. It just cares about energy output regardless of how much is useful.
ARM ISA does not offer some significant magical performance or energy eff just by itself unlike many ppl tend to believe