News RTX 5090 cable overheats to 150 degrees Celsius — Uneven current distribution likely the culprit

[SydneyAngel]

How does this happen for 2 generations in a row. 2k just to have your could melt smh

 

[_D_D]

The 150C was at the PSU end indicating that as likely the main problem. Note how Der8auer had to 'forcefully wiggle' the connectors to remove them from the PSU. Given the smaller pins and sockets used on today's modular PSU connections as well as newer GPU's, I have to wonder if some deformity is introduced by 'wiggling' creating higher resistance on some matings or/and poor design.

 

[InvalidError]

I don't think anyone even knows for sure at this point for certain if its the contact, or lack thereof, between the pin and card that's building heat or if it's an internal failure in the cards connector and PCB.

The damage pattern seems pretty clear about that: the worst of the damage is practically always on the female connector. The almost paper-thin metal tube the female connectors are made from cannot take much abuse before losing the amount of compliance and shape memory they need to maintain good enough contact across the whole connector. In some of the earlier connector failure analysis, there were some where the female pins broke off from the connector-side bus bar or the bar strip itself broke.

The male pins are just solid bronze or similar, very little can go wrong with them. That is why the most wear-prone parts of the connection are on the conveniently replaceable cable in-between.

The cable-side of the current balance problem is easy to solve: run a single #6 wire pair as a bus bar connecting the two ends to each other. No more wire-wise balance to worry about.

 

[YSCCC]

The 150C was at the PSU end indicating that as likely the main problem. Note how Der8auer had to 'forcefully wiggle' the connectors to remove them from the PSU. Given the smaller pins and sockets used on today's modular PSU connections as well as newer GPU's, I have to wonder if some deformity is introduced by 'wiggling' creating higher resistance on some matings or/and poor design.

Of course it could be, but then since it need to be tight to ensure contact, it will be very difficult if not impossibile to just pull it out. But then there's the problem, you need to be able to be secure, yet robust enought to let it survive reasonable number of cycles to work flawlessly, and that's partly why the old standard have so much headroom with the spec. And that's why there's a need for the reciever end to detect early signal of a problem and warn you to check/replace the said part, which sadly isn't present in 4000 series and 5000 series.

 

[InvalidError]

P.S. car brakes have wear indicators to warn you, a small metal is protruding when brake pads still works but is approaching dangerous level needing replacement and it will squeak like hell when you brake.

Pads are less than 1/10th of all of the things that can and likely will go wrong with brakes if you neglect them long enough or drive in conditions that may accelerate degradation. Guide pins can seize. Rust under shims can lock pads in place. Bad brake fluid can cause pistons themselves to seize. Brake lines can burst. Etc. On a car, there are dozens of maintenance items that don't have dedicated reminders.

 

[InvalidError]

The 150C was at the PSU end indicating that as likely the main problem.

Temperature is a symptom, not a cause.

He had two wires getting considerably warmer than all of the others, which strongly suggests that his cable was damaged at either or possibly both ends and only had two 12V lines still in usable shape.

 

[YSCCC]

Pads are less than 1/10th of all of the things that can and likely will go wrong with brakes if you neglect them long enough or drive in conditions that may accelerate degradation. Guide pins can seize. Rust under shims can lock pads in place. Bad brake fluid can cause pistons themselves to seize. Brake lines can burst. Etc. On a car, there are dozens of maintenance items that don't have dedicated reminders.

Yet those maintenance items won't kill you, sized pads only slows you down, not bursting into flames. And cars have specified inspection cycles.

Yet these stupid 12vhpwr connectors don't, and by no means you can inspect whether it is working or not until it is burning.

 

[InvalidError]

Yet those maintenance items won't kill you, sized pads only slows you down, not bursting into flames. And cars have specified inspection cycles.

Seized pads can cause loss of control. They can also cause the brake calipers to get hot enough to boil brake fluid, cook piston seals and ignite brake fluid/steam that ends up leaking out.

Yet these stupid 12vhpwr connectors don't, and by no means you can inspect whether it is working or not until it is burning.

You have 10 contact heat sensors on your hands. If your HPWR connector is going bad, pretty sure your fingers will detect 150C.

 

[YSCCC]

Seized pads can cause loss of control. They can also cause the brake calipers to get hot enough to boil brake fluid, cook piston seals and ignite brake fluid/steam that ends up leaking out.

completely seized pads will be immediately noticeable when you drive on the road, and if it seized in a hard braking, you are going to stop and not going on road with on side seized

You have 10 contact heat sensors on your hands. If your HPWR connector is going bad, pretty sure your fingers will detect 150C.

O yes, but the case is closed, and by the time fingers sensed it, it's damaged already. what a great and reasonable design by Nvidia

 

[InvalidError]

completely seized pads will be immediately noticeable when you drive on the road, and if it seized in a hard braking, you are going to stop and not going on road with on side seized

Getting a pad seized hard enough to lock wheels is unlikely since the rust wedge itself is resisting pads getting squeezed in but they can be seized enough to make traction recovery on snow and ice impossible. That is the kind of seized that doesn't necessarily register as a problem until it is too late.

When my father gave me his previous car and I did a general check-up, I found out he had been driving with seized pads and melted slide pin boots on both sides. I even had to beat the pads out of the caliper brackets with a sledge hammer and a crowbar, completely destroyed them by the time they came off so I could grind all of the rust off. Last year, he had a stuck pad problem on his 12-years-new car that caused a chunk of his front-left rotor to peel off from rust setting in from under the heat-affected layer. That was quite the interesting test-drive. Both of his rear calipers were also seized, probably from wherever he got his scheduled maintenance done not bleeding rear brakes based on how stuck the bleeder valves were.

 

[YSCCC]

Getting a pad seized hard enough to lock wheels is unlikely since the rust wedge itself is resisting pads getting squeezed in but they can be seized enough to make traction recovery on snow and ice impossible. That is the kind of seized that doesn't necessarily register as a problem until it is too late.

When my father gave me his previous car and I did a general check-up, I found out he had been driving with seized pads and melted slide pin boots on both sides. I even had to beat the pads out of the caliper brackets with a sledge hammer and a crowbar, completely destroyed them by the time they came off so I could grind all of the rust off. Last year, he had a stuck pad problem on his 12-years-new car that caused a chunk of his front-left rotor to peel off from rust setting in from under the heat-affected layer. That was quite the interesting test-drive. Both of his rear calipers were also seized, probably from wherever he got his scheduled maintenance done not bleeding rear brakes based on how stuck the bleeder valves were.

that is off topic but in a bit I've been owning my car (a mazda) for 16 years and went through 300k km, all these are pretty simple feel and visual drive off able to detect and go for a fix, same as the much larger 8pin connector in the GPU.

Now the issue we have on the melting gate 2.0 is that, it is NOT visual inspection-able, you can either being very wasteful to buy a new cable everytime you unplug the old one, or if you get a brand new defective one and still got your cable metled. And still they can claim it's user error "you didn't plug it in right". That isn't remotely acceptable, imagine it's your car but all brakes and rotors are hidden and there's no way you can see if there's rust or oil leak?

 

[Peksha]

Temperature is a symptom, not a cause.

He had two wires getting considerably warmer than all of the others, which strongly suggests that his cable was damaged at either or possibly both ends and only had two 12V lines still in usable shape.

You are making a false conclusion again. This shows your bias.
The only conclusion from the video is an inadequately designed solution for powering the GPU. Obviously, the people who came up with it do not know either electrical engineering or electricity

 

[Gururu]

Here come the 5080 report(s)...

 

[YSCCC]

https://www.reddit.com/r/pcmasterrace/comments/1io4a67/an_electrical_engineers_take_on_12vhpwr_and/

And some proclaimed Electrical engineers explaining his view on how it's the connector and Nvidia's fault of all these, not cable or user

 

[palladin9479]

https://www.reddit.com/r/pcmasterrace/comments/1io4a67/an_electrical_engineers_take_on_12vhpwr_and/

And some proclaimed Electrical engineers explaining his view on how it's the connector and Nvidia's fault of all these, not cable or user

Sums up the entire situation with 12VHPWR

The connector itself isn't bad. It is simply rated far too high, leaving little safety factor and thus, little room for error or imperfection. 600W should be treated as the absolute maximum power, with about 375W as a decent rated power limit.

The six and eight pin cables we are used to are overbuilt by more then 2x, it's damn near impossible for real world situations to cause them to pull too much current. Several pins could be completely dead and you'd still be well within the safety margin. Conversely the 12VHPWR connector is underbuilt with very little safety for real world situations. If any individual pin fails there isn't room for the other pins to carry the load. Now the designers of that standard thought about that and put in a sense standard to tell the device to pull less power.

Despite nVidia being one of the sponsors for the new design, they seem to have decided to completely ignore this feature with the 40 and 50 series FE cards, and AIB partners followed suite.

Now we get to the 4090 and 5090 FE boards. Both of them combine all 6 12V pins into a single block, meaning no current balancing can be done between pins or pairs of pins. It is literally impossible for the 4090 and 5090, and I assume lower cards in the lineup using this connector, to balance their load as they lack any means to track beyond full connector current.

Some partners like ASUS attempted to work around the problem by putting their own sense circuitry that while not being able to balance the load, can at least detect something isn't right and tell the card to reduce power to save itself.


In this sense the cable itself is fine, as long as we keep it to the original 375W spec and not treat it as a 600W+ spec. I'm going to be sticking with the current eight pin plugs using AMD cards. The 12VHPWR should be ok for mid range cards like the 60 or 70 models and AMD's 7900XT or future equivalent. Anything bigger needs a more capable power delivery design.

 

[DRagor]

In this sense the cable itself is fine, as long as we keep it to the original 375W spec

I will call you on this when we get first 5070TI melted and somehow I have a feeling we will see one soon enough.

 

[palladin9479]

I will call you on this when we get first 5070TI melted and somehow I have a feeling we will see one soon enough.

The 12VHPWR has plenty of headroom to safely handle 375W at system boot, but then switches to a 600W maximum after software, meaning it's relying on software to be safe. The 375W value means safe under non-ideal circumstances. What is happening is the 5090's (and 4090's before) are pulling so much power they are either right at or well past the 600W maximum and anything other then perfect has a chance of bad things happening.

Basically both the 4090 and 5090 (maybe 5080) need two 12VHPWR cables to be safe. nVidia doesn't want that since it kinda defeats the entire reason for 12VHPWR in the first place.

 

[lmcnabney]

The 150C was at the PSU end indicating that as likely the main problem. Note how Der8auer had to 'forcefully wiggle' the connectors to remove them from the PSU. Given the smaller pins and sockets used on today's modular PSU connections as well as newer GPU's, I have to wonder if some deformity is introduced by 'wiggling' creating higher resistance on some matings or/and poor design.

It could go either way. Most of the sixteen pins on that connector are for signalling between the GPU and PS. Issues with communications can trigger the PS to overdeliver current or for the GPU to pull too much from a single wire.

I stand by me statement in the previous thread. The previous 8pinx3 solution provided nine +12V power wires. The 12VHPWR connection pulls the same power through six +12V power wires. All of these wires use 16ga copper. No matter how you slice it the new connector can pull 50% more current through each wire while staying within spec. If anything goes wrong and the GPU pulls too much the connection was already delivering power right at the material's limits.

 

[_D_D]

Of course it could be, but then since it need to be tight to ensure contact, it will be very difficult if not impossibile to just pull it out.

FWIW for my modular PCIe connectors I find they usually don't come out straight away from the PSU and need a constant pulling force on the connector for about 2 or 3 seconds with the latch open before they magically pop out, no perceivable wiggle. Of course some connections may be stubborn / stuck and need a little wiggle in some situations but personally I try to avoid that.

Temperature is a symptom, not a cause.

It's both. Higher temperature causes higher resistance. However my point was that the PSU connector was substantially hotter than the GPU connector. IOW the GPU connector out performed the PSU connector.

I stand by me statement in the previous thread. The previous 8pinx3 solution provided nine +12V power wires. The 12VHPWR connection pulls the same power through six +12V power wires.

I hear you. Note pin / socket size too.

 

[SyCoREAPER]

Why good morning theorists

Let's start the morning off.

View: https://youtu.be/oB75fEt7tH0?t=-1

 

[V8VENOM]

Staying within the 12V realm, the simplest fix to horrible current balance issues is ditching the multi-pin, multi-wire setup in favor of something like XT-60.

Funny you mentioned XT-60's ... I've used those before in various applications. They are certainly better than current garbage connectors used in PCs, BUT they are solder only (no crimping) ... which is fine, but it does require quality soldering. Mil spec wire (nylon insulation) allows for much higher current and mil spec connectors WILL ensure a positive connection.

 

[SyCoREAPER]

Funny you mentioned XT-60's ... I've used those before in various applications. They are certainly better than current garbage connectors used in PCs, BUT they are solder only (no crimping) ... which is fine, but it does require quality soldering. Mil spec wire (nylon insulation) allows for much higher current and mil spec connectors WILL ensure a positive connection.

Absolutely. Id love a garden hose sized connector lol.

As for the XT-60, solution to soldering is simple, have the cable hardwired to the card, long enough for the average clean tuck to connect to a another from the PSU.

Or we can run with the even more terrible idea of running over 600w through the motherboard what that inline power slot to the PCIe slot instead of 12VHPWR

 

[SyCoREAPER]

We got another one..

https://www.laptopmag.com/laptops/gaming-laptops-pcs/another-melted-cable-nvidia-rtx-50-series

 

[InvalidError]

You are making a false conclusion again. This shows your bias.

A 20X current imbalance between conductors of a cable that is shorted at both ends is impossible unless the low-current pins/wires are heavily damaged, broken or have a major defect that should never have passed QA.

derB tried some brand-new cables on the same PSU and GPU, those were within 15% from best to worst every time.

 

[InvalidError]

What is happening is the 5090's (and 4090's before) are pulling so much power they are either right at or well past the 600W maximum and anything other then perfect has a chance of bad things happening.

Cable and PSU manufacturers have repeatedly demonstrated that their "perfect" HPWR cables can pass 1000+W fine. 600W isn't supposed to be remotely problematic unless something got damaged or has a manufacturing defect.