So OS on 1TB Storage Card will impact battery life? Good to know (or not ha-ha), now I want to get my NVMe as soon as possible
It could; it’s basically any USB device that isn’t runtime suspended could cause that. You can use tools like powertop, sensors, perf, btop, mission center (and the list goes on) to figure out what causes what for you.
1 Like
The “load average” of Ryzen AI 9 HX 370 is also unnaturally high. With the EC only draws 2.7A from the official 60W charger, the battery would be constantly discharging when fully loaded if the value is accurate. Since there’s no report of battery discharging significantly when gaming on the Ryzen AI FL13, it’s more likely measurement error. If the CPU TDPs are unchanged i.e. 15W eco, 25W balanced, 35W performance, that would indicate lower power conversion efficiency of the circuit(if the measurement is accurate).
From the reading of my USB meter, the power draw is about 46~48W at full core load in performance mode so at least the 7840U measurement is accurate
1 Like
Usb ethernet adapter indeed adds about 1W to min idle, good to know. Not sure how much of that is the adapter and how much the rest of the system. I just grabbed a cheap 2.5Gbit one.
According to these measurements for the FW13, Intel is slightly more efficient when it comes to power consumption compared to AMD. (Core Ultra 7 and Ryzen 7 7840U measurements are comparable and might be close enough to fall within the margin of error.)
Crazy how much power is consumed when idle across the board. 7.6 kWh idle minimum for Ryzen AI 9 is wild.
Those are not efficiency measurements. They are absolute power readings in different states. Efficiency is energy per unit work done and amd is extremely strong in that regard. Both the 7840U and especially the ryzen ai look anomalously high in that table especially on min idle. This table seems to be more a “how fully baked is power management at launch” which intel at least used to be a lot better at than amd though it seems that intel is also slipping there.
Mind your units but even 7.6W is way high. It may already be lower but unfortunately notebookcheck does not retest that usually.
7 hours light tasks, from 80% charge
1 Like
One aspect that makes this difficult, is the measurements taken to measure power used are not accurate.
When measuring power use accurately, one needs it to be measured in IQ. I.e. Amplitude and Phase at very short intervals, and then integrate across them to reach a “watt hour” result.
I think some of the chips in the laptop can be coaxed into allowing one to read a more accurate figure. In the FW16, the ina2xx chip might me able to give more accurate information.
That’s for alternative current (AC), laptops use direct current (DC) so there’s no phase to measure
True.
But the accuracy problem still remains. Measuring average current fed into a DC-to-DC buck/boost converter is difficult to do accurately.
The ideal is to have a sample rate that will be faster than the rate of change of the current.
We don’t have that, so what we can measure, a sample a few times a second, is not accurate.
If the sample rate is a few times a second, the best we can achieve depends on the low pass filters to the sampler.
I am using the power measurements from the battery which should be reasonably accurate on average as it needs to be for the whole coulomb counting to work.
Now I’m getting 5-7W more power draw (on idle) from iGPU on 6.15.1 :c
I’m moreso split between the 7640u and the Ultra 5 125H, which of the two can last longer? It seems the AMD chip is more tweakable (i.e. undervoltable? if I’m reading it right) per this, but how does that affect real world performance? And if it does, would it make up for the AMD model’s smaller battery?
Doing what?
Unfortunately no undervolting but you can tweak the powerlimits which apparently you can’t on the intel platforms anymore.
Text editing, coding, and light web browsing.
Also, according to this reddit post the ultra chips can have the clocks lowered, how does that compare to power limiting?
Both should be doing fine there, as long as you can keep background tasks in check,
It’s a much cruder way to do it.
Yes. Like they must have measured wrong high. Lowest I have seen from my HX370 was 3.6W (reported from the battery) with nothing else running in Windows and after letting it sit for a few minutes idling to fully calm down. This is still 1 entire Watt higher than the 2.6W my 12th gen board does with the same SSD (but when I originally measured this, I only had 32 GiB RAM installed. I don’t remember if I reconfirmed this after the upgrade to 64 GiB, which I now also have on my HX370).
But also, the AI 300 board seems very wasteful near idle. Windows stupidly wants to rescan 1000s of files after every reboot (the search indexer). This causes like < 1% CPU utilization. But enough to keep the system much more awake than it would need to, including the SSD. In this state, I actually see ~10W, which I believe is way more than what my 12th gen got (6-8W) during this.
So very likely that they just had too much stuff running in the background, because the AI 300 system is even more sensitive to this as the Intel systems.
Looking at the dynamic memory clocking and CPU frequencies, they seem to stay high most of the time (AMD quotes a minimum clock of 600 MHz for all the cores, but I never see lower than 1400MHz no matter what.
Also, it seems, idling, Windows picks one of the Zen5 cores to use, instead of one of the Zen5c cores, even though they are supposedly more power efficient (using the big cores was also the case for Intel’s older hybrid designs. But there, Intel explicitly said that the E-cores were not more power efficient, just space efficient).
Also, does anybody have numbers for what their officially supported DDR SO-DIMMs report as power consumption? Do we know how accurate those are on average (mine only seem to only have a granularity of 125mW, switching between 0.250W and 0.375W per DIMM idling (mostly the latter). I have Kingston 5600 CL40 RAM, that may cause higher consumption do to its quicker latencies. Although with dynamic latencies and clocking, there should not be much excuse to let that impact what the memory is used with during idling. But they do actually seem to be 1T modules, while most JEDEC DDR5 modules are 2T.
Not sure the c cores are supposed to be more efficient at the same clock but the full fat ones do clock higher and have more cache. The main feature of the c cores is using a lot less area so you can put more cores on a chip or the same number and use the space for random ai accelerators and what not. There are apparently actual power efficiency cores planned for zen6.
The firmware should tell windows what cores are the most efficient though. Anyway ultimately this will probably still improve over time like it did for phoenix.
The impact of ram on min idle was quite small at least on my 7840u (we are talking 0.2w for going from 4800 to 5600cl40), it may be a problem for sleep but I don’t tend to use that.
Sure, but its nowhere near what Intel did. Where 8 E cores took up less room than just 1 P core.
According to AMD, the cores are supposed to have the same IPC (same micro arch), just not focused on as high of frequencies and less cache.
4800 CL40?, so the same latency class as 5600 CL46?
Also, yes that is how it should be. Especially on idle with dynamic clocking, they should both basically end up at the same frequency and latencies.
But without AMD giving us actual memory controller support specs, unless you have the comparison yourself, its very hard to know what is normal and what is not. That is why I am double checking if it could be my component selection that makes up the difference or not.
And slight differences could mess with putting memory and memory controller into the lowest power states or using the dynamic frequency well.
Yeah intel kinda gone overboard there.
Zen5c is interesting, basically the same circuitry but with different manufacturing settings that make it not clock as high but also use less space. Of course stuff like that need good scheduling to not murder single core performance but it is nowhere near as big a minefield as the intel e-cores were. I am looking forward to the zen6 actual efficiency cores that may sacrifice clock and area for a lot of power efficiency. Having a few (or even just one) of those to pawn off low intensity background stuff to would be awesome. Intel already has something like that with the ultra e cores in the pch on some of the ultra chips but I am not sure how well software deals with another entirely different kind of core.
Whatever jdec 4800 is, no idea what the actual cl was, probably not 40 though.
Well I did do the comparison, mostly cause I got my hands on a super cheap stick of 4800 and was curious.