I wanted to share something I’ve discovered regarding using ectool to set custom limits on the PSU current (likely something Framework tried to tell us about). I’ve been overriding the current limit on my PSU to 11k or 12k so that I can get full use of the device, as when it starts up I only pull 205W from it and the battery drains while gaming. However, I’ve been noticing that after a gaming for an hour or two, the PSU becomes unstable and disconnects/resets momentarily. Yesterday, I measured the power draw from the wall (I was testing the 120v sockets on my new Anker Solix C1000 gen 2) and found that at the start of a gaming session, the power draw is 205W, however, after the brick has had a chance to heat up (note that I’ve also added heatsinks to my PSU to help with dissipation) the power draw will increase up to 240W. I haven’t yet measured if there was 240W going into the laptop, i.e. to check if that’s efficiency losses, or actual power delivery, but I thought you all should know. I suspect there is a thermal delta component to their 90% derating, and I suspect if Framework reads this post they’ll be thinking “well, yeah. I done told ya!”. Before the brick heats up, the current override works amazingly, but once it’s thermally saturated, it defintely becomes unstable IME.
@James3 pinging you because I know you pay particular attention to these bricks.
For anyone considering this on the 13, there it may make things worse as you can increase the limit just fine but that’ll also make it run into input current prochot which will make performance pretty spiky. On the 16 that is not enabled so not a problem there.
For the case of adjusting it down, just keep in mind it only sets the limit on the charge controller and won’t influence any of the other power limit calculations, those are usually based on negotiated pd values. So turning it down will put less load on the charger but probably a lot more on battery.
I just ordered some usb powered fans I plan to add onto the heatsinks I added to the PSU. I’m curious to see if that will allow the over-current mod to remain stable.
Alright, I can now confidently confirm that the ectool override instability is not a result of over-temperature. With this setup, the power brick itself stays quite cool to the touch, basically room temperature, but the same power cutouts I was experiencing at both 11k and 12k current overrides remain. I’m now forced to admit that Frameworks 90% derating is likely necessary for my device.
The problem is that in the 16 the charge controller is doing the current limiting behind the first buck converter so you can’t know for sure how much you draw from the psu. So if the efficiency of that drops over time because it warms up or the output voltage of the power supply droops (which is relatively normal behavior to signal downstream devices to reduce draw before goes into ocp) it can’t know so you have to add quite a bit of error margin to the limits there. In the case of the 13 where it is directly at the psu this is a lot less understandable (my guess is still that it is because of the input current prochot)
I’ve also noticed that if I boot up my FW 16 with it plugged in my dGPU will cap out at ~70W power draw (even though the budget is set to 100W), but if I unplug and replug it in, then the power draw goes up to the expected 100W. Does anyone know why this is? It’s almost as if the device think’s it’s plugged into a 180W PSU unless it’s reconnected while powered on.
I didn’t have this particular issue, but what I did have is power profiles daemon giving different amounts of power in the same state, e.g. if one switched from Performance to Balanced and back, it would cap at 70-80W.
I only noticed because at this power the battery wasn’t draining, but at the same time I had way more performance than with the usual 40W cap in Power Save.
I used this to, heh, balance power draw when gaming.
Hmm, perhaps this is closer to what’s happening. I do have it setup to auto switch power profiles when plugged and unplugged, but perhaps those only partially work during a fresh boot. I’ll do some testing to see if going back and forth also adjusts the power budget. IIRC, I tested this before and it does… I just don’t remember for sure.
@tade0 I did a quick check, and it doesn’t look like toggling between balanced and high-performance recreates the issue/workaround for me. It appears to only happen on boot vs replug.
Despite the cooling not appearing to make any real differences… Amazon wouldn’t cancel my order for the extra parts so I guess I’m leaning into it anyways. I cleaned up the zipties and cable management, now the cooling mod looks a bit nicer.
I think the problem is almost entirely caused by the inefficiencies of the buck+NVDC power circuit. In comparison I have a (non-Framework) gaming laptop with RTX 3070 (TGP 115W) and Ryzen 9 5900HX (TDP 65W short, 54W long, 45W sustained). I got full power when plugged in the 240W(20V12A) power supply and running GPU and CPU stress test at the same time at 5% battery left. The charging wattage was 13W(54W CPU power) then changed to 30W(CPU power reduced to 45W). Consistent performance when using power adapter regardless of the battery level. When you add the numbers up (30W+45W+115W)/0.81 you’ll get power between 230W to 240W. I used 81% is because 90% of DC-DC and 90% of VRMs efficiency for GPU and CPU, the real efficiencies are probably higher at about 92% however I did ignore other power consumptions such as the screen, RAM, peripherals, etc.
The only big difference between this and the Framework is that the latter uses 48V5A then converted to 20V. Due to the efficiency loss the current can’t achieve 12A on the 20V side. This causes both battery discharge when gaming and lower performance if the battery level is low despite using a 240W adapter.
Do the calculation again for the Framework (54W+100W)/0.81/0.9/0.9 and you’ll get almost exactly the same result (less than 1W difference). I removed the battery charging wattage(since the Framework isn’t charging when playing demanding games) added another 0.9 for additional buck converter efficiency and another 0.9 for capping to 90% current of the PD charger in the EC program
A note, I have a 240w dock that is also rated for 5A+ and used the charge param changes. I had moments where my dock/connection would reset during heavy use, but after some dock fw updates I did not see it happen again.
