PL2 limit engages when dGPU is under load, but stays engaged until reboot

[PositronCannon]

Been trying to figure this behavior out for a couple days now, but I'm just completely puzzled.

For context, this is a 2022 Acer Helios 300 PH315-55-7174, with an i7-12700H and an RTX 3060 140W (supposedly - I haven't seen it go past 120W and that's in Furmark which is an extreme example, in games it stays around 100W but I think both cases may be due to a system-wide power limit, more on that later).

Basically, if I do a CPU-only bench after a reboot, the CPU will reach around 110W no problems, and actually stay there for quite some time before slightly throttling down due to VRM thermals. So no problems on that front, and it's not a realistic scenario anyway, if anything I'm impressed the CPU itself doesn't reach its 100C thermal limit.

If I then start up a GPU benchmark (or a game, really anything that pushes the dGPU), the CPU will throttle down to 45W with PL2 as the limit reason. This doesn't surprise me as from what I've observed, the laptop seems to have a roughly 150W overall power limit for CPU+GPU (which to me seems kinda low for a 280W power brick but whatever, it's fine).

The weird part comes when I stop the GPU benchmark, at which point the CPU will go to up 56W and stay locked there, with PL2 limit still active. I figured it'd just be a temporary thing and it'd take a while of the whole system being idle for PL2 to reset, but nope, from that point on the CPU is simply capped at 56W regardless of how much time passes. The only way I've found to get it back to boosting to 110W is with a reboot.

Admittedly, in practice this isn't a huge deal as I rarely run CPU-only workloads, but... still, this can't be the intended design, right? I'm really curious if anyone else has seen this sort of behavior, and if there's anything I could change in ThrottleStop to fix this (though I'm afraid it may be something Acer botched at the EC level). I tried messing with the TPL settings yesterday such as disabling PL2 but I think I made things worse and somehow capped the CPU to 56W period, even after a reboot, so I deleted the .ini and started over from stock settings which are the ones I've been testing with now. I'm fully open to suggestions on this front.

The Windows power plan is set to the "hidden" high performance setting. Obviously, laptop is connected to AC power and battery is 100% throughout. Also, changing the performance modes on Acer's PredatorSense software from Default to Extreme or Turbo does not seem to make any appreciable difference in terms of power limits.

ThrottleStop logs of the above tests: https://pastebin.com/2eACRrvh
Some notable timestamps:
- 23:01:54: TS Bench starts, CPU at 110W (all good)
- 23:09:58: I start the GPU bench (Furmark), a few seconds later CPU goes down to 45W (probably normal)
- 23:11:13: I stop the GPU bench, CPU goes up to 56W (weird but ok)
- 23:18:55: After 6 minutes idling, I start TS Bench again, CPU still caps at 56W (???)
- Not in log, but about an hour of idling later, CPU still caps at 56W

Any ideas? And thanks to anyone who read all that!

edit: oh, and thanks to unclewebb for making this software, it really helped my old thermally challenged laptop be noticeably less thermally challenged. Too bad Intel killed undervolting for most CPUs now, so I mostly use it for monitoring on this new one...

 

[unclewebb]

intended design

Likely just poor design. If you complain, Acer might fix it by locking it down to 56W all of the time.

botched at the EC level

That would be my guess. You can try checking the MMIO Lock box. If this does not change anything then I do not know what else to try. ThrottleStop has no access to the EC power limits.

The log file looks like different teams of engineers working on different throttling schemes. No real logic to sometimes 45W, sometimes 56W and sometimes over 100W.

 

[PositronCannon]

Unfortunately the MMIO lock option didn't help, same for other settings I tried. I guess it is what it is, at least it's not a huge deal in practice as I mentioned, and it can be worked around by rebooting first if I really need max CPU power for specific tasks. But man, what a bizarre design flaw.

I'd say some of this behavior does have a logic in terms of what the GPU's current power draw is, when considered together with the apparent 150W CPU+GPU power limit. Since the GPU generally maxes out at 100W, I get limiting the CPU to 45W in this case as the math checks out. What makes much less sense is the second seemingly arbitrary 56W limit after the GPU goes idle (what exactly is this based on? and can't the CPU power limit be adjusted dynamically based on current GPU power anyway?), and what makes absolutely zero sense is this limit just being applied forever from that point on, even when the GPU has been idle for hours. Hell if I know.

Oh well, at least I'm happy with the actual gaming performance of this laptop which is the main purpose anyway. It's just quite funny to go from a laptop where the CPU couldn't even get close to its 45W TDP without thermal throttling in combined CPU+GPU scenarios... to one where the CPU can reach 110W no problem but then throttles down for no reason.

Thanks for looking into it anyway!

 

[PositronCannon]

If you complain, Acer might fix it by locking it down to 56W all of the time.

Oh man, you don't know how right you were.

As a last resort, I tried updating the BIOS to the latest version since hey, you never know. It's not like they could make it much worse, right?

Now the CPU immediately trips PL2 locking it to 56W even right after a reboot, and worse yet, now PL1 also trips after a minute or so when it didn't before, locking it to 45W even when the rest of the system is completely idle.

Boy I sure love having a CPU running at less than half the power it would be capable of, when it's 30C under its thermal throttling limit and a whole 140W under the power supply's rating... sigh. Joke's on me, I guess.

edit: as a "bonus", this update also reduced the PROCHOT limit from 100C to 92C, not that it's ever going to reach that anyway now that it's been completely kneecapped. It already barely reached 95C before at 110W, and it was the VRMs that asked for a break first.

 

[unclewebb]

I hate to say I warned you.

When a laptop is first released, some manufacturers leave many settings like the power limits wide open. Laptops are sent out for reviews and every review you ever see looks wonderful. After all of the reviews are finished and the laptops are returned, then a BIOS update comes out and locks everything down.

It is very rare to ever see a review a few months later after the BIOS has been updated. This is especially true for laptops. A very shady business practice.

Is it possible to go back to the original BIOS? Sometimes going back is locked out too. Asus keeps many BIOS versions available for their desktop boards but for laptops, many manufacturers perform magic and the original BIOS versions disappear.

 

[PositronCannon]

I found a way to revert to the oldest available BIOS on Acer's site, but no dice, the behavior is still the same as before.

With all these reboots I ended up finding out that the loss of the ability to boost up to 115W wasn't due to the BIOS update after all, it just seems random (although it may have been more common on the latest BIOS, not sure). There must be some kind of flag that's being tripped and locks the CPU into the PL2 limit, and it's not necessarily related to dGPU power draw as I thought. Or maybe it is, since I have the system set to use the Nvidia GPU exclusively and bypass the iGPU, so a random power spike on boot could cause this. It also seems random whether it will move on to PL1 (45W) after a minute or stay at 56W.

I don't know, by this point I'm out of ideas and not gonna be bothering further. I'll just have to accept that this CPU is being way underutilized and it is what it is. To be completely fair, I understand these are the Intel-recommended power limits, it's just they're absurdly conservative on this kind of laptop.

 

[PositronCannon]

Old topic and it technically isn't related to ThrottleStop in the end, but I wanted to post an update just in case anyone with the same issue stumbles upon this topic...

So Acer released a new BIOS version (1.11) for this laptop after my last tests months ago. It doesn't mention anything about thermal management or anything like that so I wasn't expecting any change, but hey, what's the worst thing that can happen? (clearly I didn't learn my lesson already)

Well, as it turned out...

The CPU is once again boosting to the 110W it's capable of, and unlike at first, it actually seems consistent (worked every time so far in 4 reboots) and from what I can tell so far, unlike before it doesn't get automatically locked to 56W/45W after running a GPU-heavy application, or even while it's running which is even more surprising (granted I haven't tested particularly power-hungry games yet).

So long story short, seems like Acer just has no idea what they're doing with their BIOS/EC settings lol. But whatever, it seems to be fixed now at least. Fingers crossed it wasn't just a fluke and it's not actually completely random again...