CPSC Demands a Recall of CableMod GPU Angled Adapters, Estimates $74.5K of Damaged Property

[R-T-B]

Using 3xPCIe cables is definitely clunky, but I also don't want a future where GPUs routinely need 450W+

No one does but we also want advancement without significant silicon advances, so something's gotta give.

There's plenty wrong with the connector. It greaty exceeds the current of the MicroFit connector as defined by Molex, the manufacturer of the MicroFit connector.

You want to quit bullshitting?

Nvm, I need to quit playing electrical pretendo-engineer. Stuff below is invalid math as discussed below.

https://www.molex.com/content/dam/molex/molex-dot-com/products/automated/en-us/productspecificationpdf/430/43045/PS-43045-001.pdf?inline

Do your own wattage math. With 18AWG wire the 12pin connector is capable of 12v*5.5a*12pins or 792W. They just need to build it to spec. Heck, even 20AWG makes it to 642W.

 

[kapone32]

No one does but we also want advancement without significant silicon advances, so something's gotta give.


You want to quit bullshitting?

https://www.molex.com/content/dam/molex/molex-dot-com/products/automated/en-us/productspecificationpdf/430/43045/PS-43045-001.pdf?inline

View attachment 334065
Do your own wattage math. With 18AWG wire the 12pin connector is capable of 12vx5.5ax12pin or 792W. They just need to build it to spec. Heck, even 20AWG makes it to 642W.

Now that is some serious Popcorn

 

[trsttte]

Do your own wattage math. With 18AWG wire the 12pin connector is capable of 12v*5.5a*12pins or 792W. They just need to build it to spec. Heck, even 20AWG makes it to 642W.

I guess you're not an electrical engineer... Only 6 six of the connectors are supplying current, the other 6 are closing the circuit and "carrying" it away.

I believe we should be using the 6 circuit value and not the 12 for 12VHPWR connector though, either way 12*6*6=432W so still much lower than the current PCIe spec is allowing, maybe amphenol figured out something molex didn't... for me personally this connector just became more of a shitshow than it already was and I hope AMD and Intel are able to weather the pressure to start using it as well until nvidia and co. figures out the obvious and they work towards a proper solution

 

[R-T-B]

I guess you're not an electrical engineer...

No, caught me, but I do well enough to pretend usually.

Unless there's an external ground somewhere you'd be right. And looking at the pinout online, you are right. My bad here. Simple but fundamental mistake yes.

The plot thickens.

 

[Vayra86]

No, caught me, but I do well enough to pretend usually.

Unless there's an external ground somewhere you'd be right. And looking at the pinout online, you are right. My bad here. Simple but fundamental mistake yes.

The plot thickens.

Glad you figured it out in the end.

 

[Chrispy_]

No one does but we also want advancement without significant silicon advances, so something's gotta give.


You want to quit bullshitting?

Nvm, I need to quit playing electrical pretendo-engineer. Stuff below is invalid math as discussed below.

https://www.molex.com/content/dam/molex/molex-dot-com/products/automated/en-us/productspecificationpdf/430/43045/PS-43045-001.pdf?inline

View attachment 334065
Do your own wattage math. With 18AWG wire the 12pin connector is capable of 12v*5.5a*12pins or 792W. They just need to build it to spec. Heck, even 20AWG makes it to 642W.

That's wire gauge you're talking about, which is irrelevant, so long as it's adequate. The problems with ATX 12VHPR aren't desoldering wires, they're bend radius and poor insertion issues causing poor pin mating, which is why the pins get hot enough to melt the housing and cause a potential ignition source via short circuit or direct combustion.

Molex officially specify 8.5A per pair of pins, and that number gets smaller the more pins you have in proximity, since the heat output needs to be taken into consideration.

So, if it was a 2-pin MicroFit, you'd get up to 8.5A per circuit, as defined by Molex themselves.

The actual specification document of relevance is this one:

https://www.molex.com/content/dam/molex/molex-dot-com/products/automated/en-us/productspecificationpdf/430/43045/PS-43045-001.pdf?inline

Page 7 gives the MicroFit safety certifications from the three main global electrical safety bodies. Of note, not a single international safety body rates MicroFit higher than 8 Amps, and already you can see a problem with that - 8A x 12V = 96A per wire pair, so the ATX 3.1 spec that defines a 600W 12VHPWR connector is already violating the IEC, UL, and CSA safety ratings for the connector alone.

Page 8 gives the important information, invalidating the "up to 8.5A" per circuit in the ATX 12VHPWR, because the 12VHPWR isn't a single circuit:

"Appropriate de-rating is required based on circuit size, ambient temperature, copper trace size on the PCB, gross heating from adjacent modules/components and other factors that influence connector performance. Wire size, insulation thickness, stranding, tin coated or bare copper, wire length & crimp quality are other factors that influence current rating.​

​

Ambient temperature right next to a hot GPU is not 24C STP as tested.
There are adjacent modules as 12VHPWR is six circuits with double adjacency in four of them.

It is my understanding that the up to 8.5A limit per pair is not applicable for 12VHPWR, and someone with a more relevant engineering degree than mine would need to work out the "Appropriate de-rating" of that absolute best-case-scenario 8.5A per pair rating that Molex themselves provide. I don't see from those specification documents how "up to" 8.5A covers the ATX 3.1 spec in any way, other than if you accept that the ATX 3.1 spec is eating into the safety margins Molex themselves have set for their own products.

Watching that video I linked earlier persuaded me to fact-check him and do my own research - I'm more than happy to be corrected, but as far as I can tell he's correct. 9.2A as per the ATX 3.1 spec is in violation of the stated MicroFit connector's electrical safety rating. If anyone more knowledgeable than me can chime in and either correct me or validate me, please do so as I'm well aware my foundation year as an engineer only covered electrical engineering at a superficial level - enough to know the basics, but also enough to know when I need a chartered electrical engineer.

 

[trsttte]

That's wire gauge you're talking about, which is irrelevant, so long as it's adequate. The problems with ATX 12VHPR aren't desoldering wires, they're bend radius and poor insertion issues causing poor pin mating, which is why the pins get hot enough to melt the housing and cause a potential ignition source via short circuit or direct combustion.

Molex officially specify 8.5A per pair of pins, and that number gets smaller the more pins you have in proximity, since the heat output needs to be taken into consideration.

So, if it was a 2-pin MicroFit, you'd get up to 8.5A per circuit, as defined by Molex themselves.

View attachment 334148

The actual specification document of relevance is this one:

https://www.molex.com/content/dam/molex/molex-dot-com/products/automated/en-us/productspecificationpdf/430/43045/PS-43045-001.pdf?inline

Page 7 gives the MicroFit safety certifications from the three main global electrical safety bodies. Of note, not a single international safety body rates MicroFit higher than 8 Amps, and already you can see a problem with that - 8A x 12V = 96A per wire pair, so the ATX 3.1 spec that defines a 600W 12VHPWR connector is already violating the IEC, UL, and CSA safety ratings for the connector alone.

Page 8 gives the important information, invalidating the "up to 8.5A" per circuit in the ATX 12VHPWR, because the 12VHPWR isn't a single circuit:

"Appropriate de-rating is required based on circuit size, ambient temperature, copper trace size on the PCB, gross heating from adjacent modules/components and other factors that influence connector performance. Wire size, insulation thickness, stranding, tin coated or bare copper, wire length & crimp quality are other factors that influence current rating.​

​

Ambient temperature right next to a hot GPU is not 24C STP as tested.
There are adjacent modules as 12VHPWR is six circuits with double adjacency in four of them.

It is my understanding that the up to 8.5A limit per pair is not applicable for 12VHPWR, and someone with a more relevant engineering degree than mine would need to work out the "Appropriate de-rating" of that absolute best-case-scenario 8.5A per pair rating that Molex themselves provide. I don't see from those specification documents how "up to" 8.5A covers the ATX 3.1 spec in any way, other than if you accept that the ATX 3.1 spec is eating into the safety margins Molex themselves have set for their own products.

Watching that video I linked earlier persuaded me to fact-check him and do my own research - I'm more than happy to be corrected, but as far as I can tell he's correct. 9.2A as per the ATX 3.1 spec is in violation of the stated MicroFit connector's electrical safety rating. If anyone more knowledgeable than me can chime in and either correct me or validate me, please do so as I'm well aware my foundation year as an engineer only covered electrical engineering at a superficial level - enough to know the basics, but also enough to know when I need a chartered electrical engineer.

We're using molex data because it's easier to find but the 12VHPWR connector is based on the Amphenol design

https://cdn.amphenol-cs.com/media/wysiwyg/files/documentation/datasheet/boardwiretoboard/bwb_minitek_pwr_cem_5_pcie.pdf

https://cdn.amphenol-cs.com/media/wysiwyg/files/documentation/gs-12-1706.pdf

Amphenol specs the connector to 9.5A, how they did that when molex is only going for 8.5A for single pair with many derates and *asterisks for bigger assemblies (like the 12vhpwr) I can't really explain, better materials and more stringent requirements on how the cable is crimped and soldered will allow some wiggle room but it's not like they can change the physics of "metal connector surrounded by plastic".

Either way the point still stands, this is a shitshow. We went from a connector that in the worst case could still supply almost double the rated power to one that's barely able to supply it's rated load in a good day. All this while increasing the price just to save a couple millimeters of board space when 3 slots gpus are defacto standard now.

You made a very simple and very good sugestion earlier, why didn't they just use a 10 pin mini fit without the redundant sense pins!?

 

[Chrispy_]

All this while increasing the price just to save a couple millimeters of board space when 3 slots gpus are defacto standard now.

This is indeed the kicker; The GPUs using it have, up until the 40-series super launches, all been triple-slot behemoths that aren't short of space to fit a connector.

The only reason to shrink it is is Nvidia's weird ultra-cramped FE PCB that's pentagonal and all kinds of awkward to fit inside the form-over-function cooler design they're going for. If it was actually a good cooler then it wouldn't always be so close to the bottom of the noise and thermals charts...

 

[R-T-B]

Glad you figured it out in the end.

I'm not allergic to facts. I actually enjoy learning. I just sometimes suffer a bit from "know-it-all" syndrome lol.

It does indeed make me question the design more, but I am still thinking the design is far from catastrophically flawed. I will admit a lot of heartache could probably have been avoided with just looser allowable tolerances though yes.

 

[Panther_Seraphin]

The datacenter/enterprise have rejected 12V on a large scale for powering high end accelerators recently.

OAM modules are 48V primarily with 12V and 3.3V for compatability/secondary purposes with a 50 watts on 12V and 10 on 3.3V. Honestly with the efficency benefits of a higher voltage and theoretical power limits that raise due to that it should have been something AMD/Nvidia/Intel had came to an agreement with a while ago.

 

[trsttte]

The datacenter/enterprise have rejected 12V on a large scale for powering high end accelerators recently.

OAM modules are 48V primarily with 12V and 3.3V for compatability/secondary purposes with a 50 watts on 12V and 10 on 3.3V. Honestly with the efficency benefits of a higher voltage and theoretical power limits that raise due to that it should have been something AMD/Nvidia/Intel had came to an agreement with a while ago.

ATX12VO was a compromise in that direction, to maintain compatibility and at the same time getting rid of secondary voltages - 5v and 3.3v - that are barely necessary for anything and hurt efficiency. Even so, everyone has been fighting, bitching and screaming on how it breaks compatibility with everything, it's unnecessary and a bad design.

It's tough, changing everything to 24v or 48v would easily improve efficiency and halve current in a way discussions like this one about how much of a disaster it has been putting micro fit connector in gpu's would be irrelevant, but everything would need to be redesigned to match it, it's been hard enough getting people on board with a small change like ATX12VO let alone if everything would need to change.

 

[Papusan]

Either way the point still stands, this is a shitshow. We went from a connector that in the worst case could still supply almost double the rated power to one that's barely able to supply it's rated load in a good day. All this while increasing the price just to save a couple millimeters of board space when 3 slots gpus are defacto standard now.

AIB parters don't need to follow Nvidia's strict guidelines in everything for the PCB and use single 12VHPWR power connector. They are all free to follow galax and double up. They are even free to add safety mechanism that cut/reduce the power once one of the fragile connectors lose signal. The only change I wished have been done is a 50% bigger 12+4 pin power connector. As well made the locking mechanism less fragile. In short a full re-design.

 

[Chrispy_]

It does indeed make me question the design more, but I am still thinking the design is far from catastrophically flawed.

It is definitely "far from catastrophically flawed", there are good things to take from the design:

• The idea of using smaller wires for sense pins is good. There's no need for sense pins to be huge or use bulky AWG16 wiring.
• The idea of adding more pins per connector is good because nobody likes having to run multiple cables and multiple connectors when one will do.

The mistakes outweigh the good though:

• It was shrunk, for seemingly no good reason - making PCBs smaller just to save $0.10 cents on fibreglass isn't smart on physically massive $600+ GPUs with huge profit margins.
• There was no need to double the current per pin over our previous PCIe standard and completely throw all of the tolerance/safety margin out of the window. Why?!!! If they need more power add more pins.

 

[Panther_Seraphin]

It's tough, changing everything to 24v or 48v would easily improve efficiency and halve current in a way discussions like this one about how much of a disaster it has been putting micro fit connector in gpu's would be irrelevant, but everything would need to be redesigned to match it, it's been hard enough getting people on board with a small change like ATX12VO let alone if everything would need to change.

I know i just know there is already quite a few areas that could tolerate a 24v input. Most GPUs for example their DrMos components tend to be 24/30V input tolerant and if the controllers tend to sit on a 5V bus that usually is provided on the card itself if it isnt taken through the PCI-e connector.