[skyeto]

Even easier to use the old standard instead of making a new solution to a problem that didn't exist in the first place (except for aesthetic reasons). The old connectors were solid and had plenty of headroom if used within spec, with the new cards you'd still just need 3-4 to be well within spec - no monitoring circuitry needed.

[arghwhat]

The PCI-SIG maximum safe rating for the old and janky 8-pin connectors was 150 watt, despite how bulky it was. The 4090 would need 4 of those (5 if you account for when it exceeds 600 watt).

A new connector was absolutely necessary, even for more reasonable GPU power levels. That the current connector is broken is a different problem, and could likely be fixed without manufacturing changes without a new overall connector design.

[skyeto]

But the new connector is still only rated for 600W (if used within spec), if we directly compare that to using 4 of the old connectors (rated at 150W each) we're spreading the same load over many more pins.

A 12VHPWR connector is using 12 pins for GND/12V.

An 8pin PCIe connector has 6 pins for GND/12V, with 4 connectors that's 24 pins total.

Assuming that the pins are fairly similarly specced, the old connector should be safer, especially when exceeding the rated power. Regardless of how you frame it, it seems like aesthetics got prioritized over safety.

[arghwhat]

More pins is not safer, but just increases the risk of a single failed pin without mitigating the possible damage. Pins with good contact are safer than those without.

Pulling 50A through a connector of that size or smaller is in itself not an issue. As mentioned elsewhere, powerpole is smaller than an 8-pin connector and handles 55A at 10000 cycles without a hitch (even hot plug at 30A for hundreds of cycles). That this connector only does 40 cycles before breaking is caused only by it being poorly made or a poor design (e.g. relying on interference fit of contacts rather than true spring loading), and has nothing at all to do with the size.

The current plug is an ugly and cheap connector, so I don’t see aesthetics as a valid argument.

[Majromax]

> But the new connector is still only rated for 600W (if used within spec),

I think that's the problem right there: running the connection at its rated limit. It implies that a damaged or slightly out-of-spec connector could fail, as we see happened here.

Defensive engineering involves safety margins for safety-relevant specifications. Running a digital signal at the limits of its specification is one thing; data corruption is a detectable problem and not inherently unsafe. Running power connections at rated limits is something else entirely because the failure modes are physical damage, fire, and electrical arcing.

600W of power delivery is no joke, and I'd like to see that power carried over cabling and connectors with absolute maximum ratings much greater than that, say 1kW. Per the specs, it seems that each of the 12 current-carrying conductors in the connector runs about 9A. A dodgy connection that results in just 1-ohm of resistance would then represent 80W lost as heat at that point.

Frankly, I think that running 600W over a single 12V multi-core cable is insane. You're talking 50A of current, or about the same current rating as the largest typical home circuits for an electric kitchen appliance or car charger. In the home, that calls for thick structural cabling, but it appears (https://www.cable-sleeving.com/pcie5-12vhpwr-16pin-connector) that the in-case recommendations are to use 16AWG wire or even a touch thinner.

[arghwhat]

The power pins are in the spec rated for 9.2A each, for a total of 662.4W@12V. To be compliant the connector must handle at least that. At the same time, the max advertised capacity is 600W, so a device must pull at most that.

So by the spec it must have at 10% headroom - and to live up to that, manufacturers must add their own margin on top. Rating for 1kW would be a 66% margin and a humongous waste of copper, and it wouldn't solve this contact issue in the first place.

Regarding wire gauge - unrelated to the problem at hand - 16AWG is already overkill. For house installation in EU, we rate that gauge for 13A to 18.5A depending on thermal conditons, with 13A being "very poor" thermal conditions (no cooling and surrounded by other hot things). With 6 pairs in parallel, that gives a conservative minimum of 78A capacity - and with the excellent thermal conditions of a case with fans ("very good" is normally just a dangling cable, not one with fans cooling it), the realistic capacity should exceed 120A. As you say, we need only 50A, so that's a 140% margin! Even 20AWG would be plenty here.

[thescriptkiddie]

Another thing is that the remaining 2 pins on an 8 pin connector are sense pins. They are connected to +12V and GND on the load side, and to a voltmeter on the supply side. This allows the supply to measure the voltage drop which indicates a bad connection or overloaded wire. The new 12 pin connector does not support this.

[arghwhat]

This is not true. Both 6 and 8 pin connectors have "sense" pins (sense0 and sense1 - one in the 6-pin section, one in the extra two pins), but they are not a voltage feedback line. They serve to signal that 0W (nothing), 75W (6-pin) or 150W (8-pin) has been connected.

This is directly equivalent to the two sense pins on the new connector - except that rather than signalling 0, 75 or 150W, it signals 150W, 300W, 450W and 600W. Cable present has been given its own pin, together with a new "power stable" pin - all in the small 4-pin sub-connector, rather than wasting power pins for sensing.

No capabilities have been lost. Bad contact issues aside, it's a big improvement in every aspect.

[bioemerl]

Then when they redesign the spec they should have made the connector they used larger.

There's a problem with running 600 Watts through that connector, if this fire thing is regular issue it's going to burn someone's house down or kill someone

[arghwhat]

They could have made it even smaller. The problem is not at all the size, it’s just bad contact.

[bioemerl]

Bad contact shouldn't be an issue. You should have to be hilariously and obviously bad with the contact in order for it to be a problem. If slightly bent cables are the problem, it's clear that your cable is rated far too high for what it is and you need to mandate larger conductors or more conductors carrying less current so that there is no chance it burns unless the cable is very badly damaged or misused.

[arghwhat]

Bad contact shouldn't be an issue, but as the connector is overheating it is an issue. The only challenge in making a power connector is guaranteeing strong enough contact on every use. It could be caused by all sorts of things like manufacturing tolerance or exact springiness of the material of the pins.

The issue bending the cable near the connector is that it causes the connector pins to deflect, making the flimsy contact worse.

More conductor is unnecessary, it just needs to make a more reliable contact.

[hulitu]

Old bad, new better.