I have a question, can you manually adjust some settings to get max TDP(and max drain) on battery and min TDP on 180W PD. It would be fun to benchmark this
I’ve finally got more than 30W out of APU. The trick was to bump APU skin temperature limit. Now the sustained limit is 33-34W and it’s bottlenecked on APU hitting 100C at room temp of 28C.
At room temp of 20C it should hit around 37W sustained. It’s still quite far from 45W in the spec.
#!/bin/bash
ryzenadj -a ${1}000 -b ${1}000 -c ${1}000 --apu-skin-temp 70
Well as it seems some batches have a troubled Coolingsystem. My 7840hs also only hit around 37-40w sustained. I got an RMA and modified my new Board and now i can get around 52-54w sustained out of the Franework Cooling Solution and 7840hs
I wonder if that is a less violent solution to the stapm throttling than my smokeless one, though 37W sustained is still a lot less than the 45 my liquid metalled 13 does which is not what I’d expect from a much bigger machine also running liquid metal.
wait what? Can you increase the TDP of FL13 to 45Watts?
I can increase it to a lot more than that, but the cooling can do about 45 sustained with liquid metal, the back elevated and the fan maxed out.
Did you see any performance improvements in a benchmark as a result of this?
Can you use smokeless to undervolt your 13? To get better performance assuming TDP doesn’t change during the process
Nope, undervolting got killed on mobile platforms, probably thanks to an overreaction to plundervolt.
I would if I could but I also think especially with the U they already run them pretty efficiently, the perf/w of those cores is pretty amazing to begin with.
The Steam Deck got official undervolting support in the BIOS, but mine wasn’t stable at even the lightest setting… which makes me wonder if I’m unlucky, or if we’re not really missing out on much by losing the feature.
The steamdeck has one of the last generations that still allow undervolting if the device manufacturer chooses not to lock it down.
From what I see online most steamdecks do have some headroom so you may have lost the silicon lottery there.
On intel 5th-9th gen mobile processors it usually made a huge difference, don’t have any personal experience with ryzen mobile chips or intel ones past that.
1 Like
Ah, I see. Come to think of it, I’d be interested in checking my FW 16’s lottery ticket. I presume it doesn’t support undervolting either then.
There was some rumblings that the 7940 would support it but I have not seen anyone try it on a framework.
But on a positive note those 7x40 class chips are freakishly efficient as it is, the my still have a bunch of rough edges concerning turning stuff on too aggressively but the perf/w of the cores themselves is pretty amazing.
1 Like
I followed your FL13 steps for smokeless umaf - such as disabling STAPM, STT, etc - I still couldnt get more than 46-48 watts at 90c with the FL16.
Have you done any tests with the 16?
Not sure the settings work the same on that platform but asuming it worked those are kind of disapointing numbers for a laptop that much bigger and coming with lm from factory. Did you set a 90c temp limit or is there power throttling going on? What does ryzenadj say about your power limits.
I don’t have a 16 and am not planning to get one, it’s really cool and doing something cursed with the expansion bay certainly has a ring to it but I don’t really consider anything above 14" a laptop XD
Edit: you did up your power limits in software too right? The ec will override the soft limits whenever you reboot or unplug/replug a power supply and stuff, the smokeless stupp mostly raises the absolute limits and disables skin temp throttling it doesn’t affect the soft limits the ec sets.
1 Like
So what I found was it didnt matter if I had stapm or stt disabled. The results were the same. I set tctl to 100 and it throttled at the same point with settings off or on.
Well… kinda. Did a cinebench run today and I got it ramp up to 50w sustained (peaks in the low 60s) and stay at around 99-100… but that was with all the stock bios settings.
For whatever reason, it just wants to stay in the 46-50 range on an all core load.
Again you did set the tdp and stuff in software too and it stuck?
If you mean 99-100C for only 50W sustained that’s pretty sad, something wrong with your cooling or you in a 50C room?
Found the same for my FW 16 (stock, no smokeless).
I found an 8-thread workload (stress -c 8) to be the most power-hungry, with a genuine 45W spike at the very beginning, before dropping to a sustained 37-42W (oscillating) - peaking at temps of 93°C.
A 4-thread workload (stress -c 4) is notable because it quickly thermal-throttles at 100°C, though I don’t see it pulling as much wattage anyway.
I tested 6,7,9 as well (with 6 also thermal throttling, 7 barely not) but these were the extremes.
Note for these tests I had to unplug my 65W charger to use the battery alone because in the current firmware, there is a bug locking PPT to 30W with a charger below 80W (see Framework Laptop 16 Ryzen 7040 BIOS 3.03 Release and Driver Bundle - #137 by Seneral)
This would explain why my 60W external battery pulls less juice when gaming. Interestingly enough, the difference in power draw from a ~25W GPU draw and a ~40W GPU draw was 0 fps in Overwatch. Haven’t wrapped my head around that one yet.
Not sure about the GPU, but I have noticed the CPU using unreasonable amounts of power, too, for trivial things. I guess some microcode deciding it’s a priority to boost as high as it can for a specific task when in reality 10W would’ve sufficed.
I initially set out to fix that firmware bug myself but after realising I don’t WANT more than 30W (since the performance is largely the same) I dropped that. I just don’t trust the CPU governor to use it’s power sparingly and only when appropriate.