RPI 5 external power M.2 Hat with Metis 2

Ah, interesting issue. From what I can find, the Geekworm X1001 can only deliver max 5W via the FFC PCIe ribbon. And if I remember correctly, the PRi5 requires USB-PD negotiation to recognise a 5A supply, otherwise it defaults to assuming 3A (that needs double checking, mind you).

As you say, and initial current surge could briefly top 5A, and I’m also not sure if there’s a diode in place to prevent any potential reverse current. I have a suspicion there wouldn’t be…

I think you can add someting like PSU_MAX_CURRENT=5000 to the Pi’s eeprom config that ensures it defaults to 5A, rather than 3A. Maybe that’d get you over that initial boot hurdle, but please do check that out in some RPi docs beforehand, just to be sure it does as we’re hoping. Keep me posted!

Here are my testing steps for powering the RPI 5 and Metis M2 with a bench power supply attached to the 5V GPIO pin:

1. I tried powering just the RPI 5 using the 5V GPIO pin connected to a bench power supply using 5V,5A. However, the RPI 5 did not boot and the RPI 5 red LED stayed on.
2. I edited the EEPROM config (thank you for your suggestion Spanner), rebooted the RPI 5, and verified that PSU_MAX_CURRENT=5000 was correctly set. 
3. I was then able to power on just the RPI 5 using the 5V GPIO pin and my bench power supply set to 5V,5A. 
4. I then attached the Metis M2 to the RPI 5 using the Seeed Studio PI M2 Hat (https://www.seeedstudio.com/PCIe-to-M-2-hat-for-Raspberry-Pi-5-p-5974.html) again using the 5V GPIO pin and bench power supply set to 5V,5A.
   1. The Seeed Studio M2 Hat connects the M2 Hat to the RPI 5 5V GPIO pin using POGO pins
5. The current went up to ~150mA but dropped to 0mA and the RPI 5 did not turn on correctly.
6. I disconnected the bench power supply and tried the USB-C PD and was able to start docker and run the inference demo on the Metis M2 without issue.

Perhaps the RPI 5 cannot source enough current through the 5V GPIO pin when the Metis M2 is connected? It doesn't sound quite correct but I'm not sure what the reason could be.

I also have some more information regarding powering the RPI 5 with the USB-C PD and the M2 Hat with a bench power supply. I messaged back and forth with the Geekworm sales staff and was told the following:

Myself:
Ah, could you connect it like this instead:
USB C Power Supply -> RPI 5 -> PCIe Cable -> M2 Hat <- External 5V Connector <- Benchtop Power Supply
I’m using a RPI 5 Hat that already connects the M2 Hat to the 5V RPI 5 pins but I’m not getting enough current. My intention is to connect the RPI 5 to the USB C power supply and the M2 Hat to a separate benchtop power supply. I hope this clarifies my question, thank you for responding!

Geekworm:

We have consulted with our engineer. The XH2.54 2-pin power connector does not support an external power supply. It is only intended to connect to the Raspberry Pi 5 5V GPIO pins.
Thank you for your understanding.

Similarly, the Pineberry Pi's HatDrive! Bottom NVMe HAT also is not supposed to be powered by an external power supply (https://github.com/geerlingguy/raspberry-pi-pcie-devices/issues/559):

the power connector on the bottom board is not meant to be used with an external power supply, it is meant to be wired to the 5V GPIO pins. We are removing it from a future revision - it is not necessary and we put it there before we had our hands on the first pi5

In essence, I have not seen an M2 hat that can officially support an external power supply while the RPI 5 is powered with the USB-C PD. I am still tempted to use both the USB-C PD along with an external power supply connected to an M2 Hat but need to do more research before I feel comfortable wiring it up and trying.

​@Spanner do you have an estimate for the current draw that the Metis M2 would require when running yolov8l-pose at max MVM utilization?

Yo ​@juliannarvaez !

Hmm, I don't have exact figures for different models, but typical M.2 consumption is 5-9W. At max MVM utilisation with a large model like yolov8l, we could potentially expect it to hit the upper end of that range, potentially hitting peak power at times.

In truth, power delivery isn’t something we’ve had many problems with on the RPi, as long as the official 27W USB-C PSU is in play. Which makes me wonder if the HAT has any limitations? I don’t suppose you’ve got access to the official Raspberry Pi M.2 HAT to test this out with? Process of elimination, like.

I suppose I didn’t really realize that limiting the MVM utilization would not drop performance significantly. I actually ordered the official Raspberry Pi M.2 HAT just today. I’ll add some comparisons to the other HATs once it arrives.

Tangentially related, I did not see a difference in the MVM utilization triggering the Raspberry Pi low voltage warning when powering the chip using the PCIe cable and the 5V GPIO.