Hacker News Reader

Comment by jacknews | Hacker News Reader

  Another limit, now that I look, is the amount of power available to a PCIe card

CPUs can draw 200W or so which already seems extreme for a small rectangle of a few centimeters. But also consider that they run at low voltage, typically 1.1-1.2V or whatever. That means those tiny BGA/wires are carrying 100+ amps, similar to what a car battery delivers on starting. PCIe tracks are much bigger.

View on HN

Also most external GPUs have a direct line to the power supply outside of the PCIe slot. IDK how much if any power they actually draw via PCIe.

75 watts

Right! e.g. for RTX 4080, TGP is 320W. Discrete cards definitely have a bigger power envelope.

I measured that some years ago. It was ~60W or so pulled through the slot for a card with two extra cables running to it. Specifically I measured the current down the wires and subtraction gave me that as a reasonable estimate for the slot. Won't be perfect but pretty sure it's non-zero through the slot.

You can use GPU-Z to show the pcie slot power draw vs the PSU cable power draw

The distance that this kind of amperage has to flow through is very small, meaning that the resistance along the path isn't that bad.

PCIe traces however run 10x (or more) the distance, so heat dissipation is more of a concern there.

The distance is larger than you think, but it is spread over a lot of pins. Most of the pins on a CPU are for power and ground.

> The distance that this kind of amperage has to flow through is very small, meaning that the resistance along the path isn't that bad.

The formula for resistive wattage dissipation does not include "distance" in it.

The formula is i^2r (i squared times r) where i is current and r is resistance.

So 10 amps flowing through 2 ohms dissipates 200 watts, and it does not matter if the "2 ohms" is 1mm long or 1m long.

So what matters is the net resistance of that short distance, not the distance itself.

Reply on news.ycombinator.com