Have some good news to report - after suffering choppy frames and competing CPU/GPU power budget during gaming for some days, now I figured out a solution thanks to some advices from handheld community
I use gamemode to run the steam games and set the game mode to pin E-cores only. (I tried to park P-cores too but the result is not great and apparently there is a reason for that)
Basically on X7 358H core 0-3 are P-cores, 4-11 are E-cores and 12-15 are LPE-cores. Pinning games to use E-cores will maximize cpu efficiency and reduce switching between hybrid cores.
Besides modest fps gain (maybe 15%) game feels so much more smooth and playable. I recommend trying this approach & apparently handheld users have been doing this for years.
At least on the amd ones you get close to full power limits with that so probably not it.
So in hwinfo64 you donât have more than 35w in the max column at all?
Back when I was playing with power limits on my 7840u, that thing could hold over 85W for a couple seconds when starting from cold so thermal throttling so fast it can never reach past 40W feels a bit weird.
max is like 45W in geekbench and thatâs for just an instant before quickly throttling
Cinebench is a comparatively long test. if it doesnât hold above 35w for a few minutes I guess itâs safe to say PL2 is not affecting the benchmark result.
Grab a 100w charger, even if it isnât utilizing full power on the 65w with out the 100w negotiation it isnât going to pull as much power as it could.
Systems76s pangolin line has worse performance on a 65w charger even when it shouldnât actually need 100w.
Since the Intel is prone to throttle more than AMD when the battery is low or the charger is not powerful. True PL2 might be only achievable when using a 140W charger
Minor improvement in ST, still not reaching 15k MT.
As I said the issue is obviously thermal throttling as under CPU load it canât even sustain the PL1 limit. Iâll add some more observations here:
During single-core benchmarks e.g. geekbench, package temp will reach 90C+ within seconds
When MT load reaches 45W package power it fell to <35W within seconds, due to thermal throttling being triggered
CB2026 MT runs for minutes while GB MT test runs for just seconds for each item with breaks in between. GB is already doing FW13 a favor. At least the power spikes in GB is able to sway the result noticeably. Thereâs no point to rerun CB2026 with a high-powered charger for this reason.
They were on a larger process node so they need higher power to achieve same frequency.
However this also makes the hotspot on the chip larger, which means easier to cool than AMD chips isopower
I think now we have the reverse. Isopower Intel 18A is likely harder to cool because of reduced hotspot area. If the chip is hitting thermal limit no amount of power supplied will make it go faster.
I think FW should consider using a larger heatsink block, use a new fan with increased airflow or even move to vapor chamber in their next intel mainboard.
You can see the contact area is smaller which demands thinner heat pipes, but the total area/volume of the heatsink is much larger to accomodate higher thermal capacity.
I tried with that graphite pad on my previous gaming laptop and it broken down after some time, wonât be risking it again
Use s-tui stress or furmark CPU burner. Geekbenchâs mixed load doesnât always draw max power. I got 49W at most for about 1 second
Did you get 99C during âthermal throttlingâ if not then itâs not thermal throttling but lowered PL2
EDIT: just noticed that the ISL9241âs max input voltage is 23.4V, which is lower than 28V. The FL13P supports 140W charger which means it likely uses an additional buck converter at the upstream of the NVDC charger. This brings the same problems from FL16, notably reduced efficiency(from power adapterâs input energy), to FL13P.
My worst scenario guess (which isnât quite true, needs further testing). When using the battery, the batteryâs discharge rate is lower than total power draw (could exceed 80W if the CPU draws 60W), so PL2 is lowered. When using AC, the additional buck converter lowers the efficiency, causing less power(and more heat) delivered to the main system bus at a given power input, making the computer heat up(and probably thermal throttle) faster, PL2 is also lowered.
Edit2: For reference, I got my 7840U the exact 31W PL1(35WPL2) when I discharged my battery to 5% then connected to a 60W charger before stress testing. Charging to 29% (still plugged 60W) then it went back to 35WPL1 53WPL2
Could be a combination of EC programming and the power management issue (caused by buck+buckboost) similar to FL16. Edit: could be also the smaller batteryâs lower pulse discharge power if older chassis is used with the Pro mainboard.
Package power is power consumed by the CPU die, so itâs CPU + iGPU since the iGPU is on the CPU die. âpsysâ is system power, which varies between computer brands, some models are VSYS power(CPU, RAM, SSD, keyboard, touchpad, USB device, screen, etc) others may include battery power as well.
I started with s-tui and then also did prime95 & furmark on Windows. You can see the results in the original thread.
And yes, hwinfo shows 99C max and max value of thermal throttling as âyesâ, and also the âdynamic PL1â is lower than PL1.
As I said this shouldnât be a power supply issue.
Package power reported by s-tui is definitely not CPU+iGPU. In games it doesnât even reach 25W despite the fans on max speed and playing a video with hardware decoding consumes only 1W in package power.
I read the original thread and you were right, the PL2 and single core test were indeed mentioned thermal throttling but I canât find the exact reply. If sustained power can reach 35W it could be the heat conduction? Try using a custom fan curve to increase fan speed on single core load, since the original fan curve is too slow on single core load. However, if the advertised 60W PL2 isnât achievable without custom tuning is it really within spec?
I am pretty sure they did not go with the whole dc-dc converter in front of the charge controller thing this time since drop in replacements for the ISL9241 exist that natively support 140W pd.
