new laptop's charger inadequate for GIMPS load?

[xx005fs]

My laptop had similar issue before when it would just refuse to run at full clockspeed while charging (maybe the charger isn't providing enough power over USB-C I actually don't know. My laptop only supports USB-C charging). Then I used intel xtu to tune down my clockspeed from 3.7GHz to 3.3GHz while also applying a -0.150V offset to the voltage, which in turn dropped my power consumption from over 55W when it's actually running at 3.7GHz stock voltage to like 31W on 3.3GHz, losing only about 5% of performance. The same can be applied to GPUs but nvidia GPUs imo don't have much room to undervolt compared to AMD GPU (I dunno if those gaming laptops are power limited or not but my business laptops definitely are power limited when running both GPU and CPU load simultaneously).

Just as my past experience undervolting numerous laptop processors it's common to get a -0.08V to -0.1V offset on 4th or 5th gen HQ chips, -0.120V on 6th gen HQ chips, -0.15V to -0.20V on 7th gen HQ or G chips, and about -0.06V to -0.08V on U-series, and -0.100 for hexacore 8th gen chips. All of which would drop the power consumption by a decent amount and hopefully mitigating the issue.

[Mark Rose]

Originally Posted by xx005fs

The same can be applied to GPUs but nvidia GPUs imo don't have much room to undervolt compared to AMD GPU (I dunno if those gaming laptops are power limited or not but my business laptops definitely are power limited when running both GPU and CPU load simultaneously).

Reducing the power limit on Nvidia GPUs works well enough in my experience.

[xx005fs]

Originally Posted by Mark Rose

Reducing the power limit on Nvidia GPUs works well enough in my experience.

It surely does. But my point is that nvidia doesn't offer the flexibility as AMD GPU would as GPU Boost will usually make the frequency bounce around and that's sorta annoying imo.

[nomead]

Originally Posted by xx005fs

It surely does. But my point is that nvidia doesn't offer the flexibility as AMD GPU would as GPU Boost will usually make the frequency bounce around and that's sorta annoying imo.

On new enough NVidia cards, you can also use "nvidia-smi -lgc" to lock the clock frequency. The instruction sheet says Volta+ so it's probably only that and Turing, There's also "nvidia-smi -ac" as in application clock but that doesn't work on my GPUs so I have no idea how it works in practice. As long as the power limit isn't exceeded, the clock frequency will stay where you put it. Also if it's too high, there are voltage limits that prevent the clock frequency from going as high as you want (VRel and VOp in GPU-Z monitoring). Granted, there is no direct way for adjusting the voltage and I think that voltage offset can only adjust it up a bit (if it even does much at all nowadays).

In the power limited scenario, the card will run as fast as it can while staying under power and voltage limits. Even on a steady GPU load, the clock frequency will vary one or two steps as the boost algorithm constantly tries to find the limit, maybe goes a bit over it and then backs off again. But annoying is all that it is, there is no technical reason for not using it. And naturally, when the GPU chip starts getting hotter, the transistor leakage currents will rise, and power consumption at any given frequency will also rise.

When running with a locked clock, power usage will vary even according to ambient temperature and cooling efficiency. Of course different loads will generate wildly different amounts of heat. For example, mfaktc will run hotter than CUDALucas, because even though the latter uses much more memory bandwidth, the cores are barely working due to 1:32 FP64 in the cores. So it can take a lot of fiddling to keep the card running at optimum rate. In the long run, I've found that using just the power limit extracts more computation power with less effort on my part. And I just ignore the bouncing frequency.