Mr. Tact wrote on Feb 12, 2025, 17:02:

FloorPie wrote on Feb 12, 2025, 16:13:
As for PC-SIG standards, this is on Nvidia. They are in the drivers seat for this, you can't blame it all on the standards guys or any other company like Intel or AMD that has input. Tossing out the fanboy line doesn't help anyone. This is bad cable design, bad quality and more. They want to put 600w through one cable and they aren't over-spec'ing the cable. The whole thing is just insanity and companies being cheap all around. I don't think I'd ever see such a reputable company like Seasonic Power Supplies tell people to use a heat gun/hair dryer to warm up a wire to bend it no more than 20 degrees. Just a total janky sideshow of a gpu industry right now for so many reasons.

First you say it is on Nvidia, then you say it is a bad cable design. Which is it?

The cable is fine. The problem is as you said, Nvidia's design allows all 600w to go through one connector. The connector is not designed for that and it shouldn't happen. Which, as you correctly said, "... is on Nvidia."

Uh, its both? They've revised it what, once or twice already? It has been a right mess all around from all sides and in the end I put most blame on Nvidia since it is their cards, their oversight, bad power cable quality along with bad standards design.

FloorPie wrote on Feb 12, 2025, 16:13:
As for PC-SIG standards, this is on Nvidia. They are in the drivers seat for this, you can't blame it all on the standards guys or any other company like Intel or AMD that has input. Tossing out the fanboy line doesn't help anyone. This is bad cable design, bad quality and more. They want to put 600w through one cable and they aren't over-spec'ing the cable. The whole thing is just insanity and companies being cheap all around. I don't think I'd ever see such a reputable company like Seasonic Power Supplies tell people to use a heat gun/hair dryer to warm up a wire to bend it no more than 20 degrees. Just a total janky sideshow of a gpu industry right now for so many reasons.

First you say it is on Nvidia, then you say it is a bad cable design. Which is it?

The cable is fine. The problem is as you said, Nvidia's design allows all 600w to go through one connector. The connector is not designed for that and it shouldn't happen. Which, as you correctly said, "... is on Nvidia."

Sgt. Rock wrote on Feb 12, 2025, 10:14:
The second video from Builzoid is brutal. explains the whole issue. Nvidia cutting corners on 4000 en 5000 series cards...

Buildzoid is a straight shooter most of the time. Most of it goes over my head but generally very informative stuff on motherboards, gpus etc..

As for PC-SIG standards, this is on Nvidia. They are in the drivers seat for this, you can't blame it all on the standards guys or any other company like Intel or AMD that has input. Tossing out the fanboy line doesn't help anyone. This is bad cable design, bad quality and more. They want to put 600w through one cable and they aren't over-spec'ing the cable. The whole thing is just insanity and companies being cheap all around. I don't think I'd ever see such a reputable company like Seasonic Power Supplies tell people to use a heat gun/hair dryer to warm up a wire to bend it no more than 20 degrees. Just a total janky sideshow of a gpu industry right now for so many reasons.

The second video from Builzoid is brutal. explains the whole issue. Nvidia cutting corners on 4000 en 5000 series cards...

Ravenus wrote on Feb 12, 2025, 02:56:
I can't claim to know Jack-S about this, but a tech-nerd friend with whom I was casually discussing the matter had this to say:

It's the connector. They are passing 600 watts from a connector that's only good for 100. These idiots are going to burn entire cities down with their new connectors designed in collaboration with fellow idiots at PCI-SIG who also cannot read data sheets. High wattage silicon is a clear sign of a monopoly deciding against spending on R&D and instead spending a few cents extra on aluminium to compensate for shit IPC gains with raw electricity and heat dissipation. The tech equivalent of throwing more people at the job instead of using technology to optimize things.

It isn't that simple. Yes, that's what ends up happening. But it happens because there are multiple connectors which are supposed to share the load but the Nvidia build does not force the card to pull from all three connectors. So, if one of the connectors is "not connected" -- it simply pulls the full load from the other two. If two are "not connected" it pulls the entire power load from one connector. This seems to be the core of the problem given the information provided in the second video. If the card forced a draw from all three connectors, and shutdown if it could not draw from one of the connectors. Then the connectors would be absolutely fine.

This comment was edited on Feb 12, 2025, 11:48.

Sho wrote on Feb 11, 2025, 19:12:

RogueSix wrote on Feb 11, 2025, 16:46:
I'm no electrical engineer either but current = Volt. As the name 12VHPWR implies, this connector is made for 12V which isn't really "lots of current".

I'm really sorry, and I promise this isn't nitpicking either, but this is completely wrong. Voltage and current are not the same, and 12V is not a figure of current.

In fact, this problem would also be eliminated by moving to 24V or 48V, which have much lower resistive loss and wouldn't heat up to this degree.

The issue here is really simple. This is a 600W+ card that's pushing close to the spec envelope to the connector, and as a result has a very low safety margin. The electrics on the card aren't designed to mitigate the situation. As a result the probabilty of things going wrong is reasonably high. I'd say the bottom line is somewhere on the spectrum between poor design and negligence. Given it's a consumer product and should not require training, I think it needs addressing indeed.

The best way to address it would be backing down from this high power draw and having more efficient GPUs. The alternative would be switching to a higher-voltage bus to increase the margins, if ATX truly wants to support 600-800w expansion boards.

(If you're curious, this is also how USB-Cs newer PD modes like the 240w one get away with such a small conductor: It's 48V.)

Thank you. I try not to respond to individuals anymore, but that incorrect claim was driving me nuts!

I will add that heat is the obvious problem, so you can narrow it down to basically 2 things 1) High resistance connection - if the plug is not secure, or if dissimilar metal construction is used (different metals expand and contract differently, potentially creating larger air gaps) a higher resistance will occur. E=IR² means that if a high resistance connection results in "R" (resistance) going up, while "E" (voltage) remains the same, "I" (current) must increase dramatically to obtain the required wattage. Resistance, in layman's terms, equals heat.
2) Improper power distribution. Even with the tightest connections imaginable, a #14 THHC wire is going to melt if you try to run a city's power draw through it. Duh! 🙂 The PSU is not melting any other parts so obviously the issue is at the graphics card.

The dude was even using a clamp-on ammeter to verify and got around 15 amps. Depending on actual wire size, that might equal or exceed wire ampacity. It's too simplistic by far to simply label this as "user error" without looking into it further. (Edit: Thermal imaging clearly showed that the single hottest point was at the connector, so that's a good place to start imo. I used to do thermal imaging on all of our motor starters to find loose/high resistance connections exactly like this)

This comment was edited on Feb 12, 2025, 03:54.

I can't claim to know Jack-S about this, but a tech-nerd friend with whom I was casually discussing the matter had this to say:

It's the connector. They are passing 600 watts from a connector that's only good for 100. These idiots are going to burn entire cities down with their new connectors designed in collaboration with fellow idiots at PCI-SIG who also cannot read data sheets. High wattage silicon is a clear sign of a monopoly deciding against spending on R&D and instead spending a few cents extra on aluminium to compensate for shit IPC gains with raw electricity and heat dissipation. The tech equivalent of throwing more people at the job instead of using technology to optimize things.

Sho wrote on Feb 11, 2025, 20:52:

RogueSix wrote on Feb 11, 2025, 20:37:

Thanks for the explanations and sorry about my faux-pas (duh). English isn't my 1st language and I thought that "current" is the word for (German) "Spannung" but "Spannung" is actually "voltage".

I'm aware of the P = U * I formula which is Power = Voltage * Current/Amperage. I mistakenly assumed previously that the English word for 'U' is current when it is actually really simple as the correct English term is "voltage" and the unit is "Volt". Mea culpa .

As a fellow German speaker I get it

Wie geht es dir? I'm not bad with German as long as they speak slowly and clearly, like if I'm a mentally impaired child. My experience listening to most Germans speak in Germany was one of constant consternation because they speak so fast. I'm still set on learning it - and Italian - however. I really like Germany a lot and look forward to returning and exploring more of it. I could easily live in Berlin or Nuremberg.

RogueSix wrote on Feb 11, 2025, 20:37:

Thanks for the explanations and sorry about my faux-pas (duh). English isn't my 1st language and I thought that "current" is the word for (German) "Spannung" but "Spannung" is actually "voltage".

I'm aware of the P = U * I formula which is Power = Voltage * Current/Amperage. I mistakenly assumed previously that the English word for 'U' is current when it is actually really simple as the correct English term is "voltage" and the unit is "Volt". Mea culpa .

As a fellow German speaker I get it

Sho wrote on Feb 11, 2025, 19:12:

RogueSix wrote on Feb 11, 2025, 16:46:
I'm no electrical engineer either but current = Volt. As the name 12VHPWR implies, this connector is made for 12V which isn't really "lots of current".

I'm really sorry, and I promise this isn't nitpicking either, but this is completely wrong. Voltage and current are not the same, and 12V is not a figure of current.

In fact, this problem would also be eliminated by moving to 24V or 48V, which have much lower resistive loss and wouldn't heat up to this degree.

The issue here is really simple. This is a 600W+ card that's pushing close to the spec envelope to the connector, and as a result has a very low safety margin. The electrics on the card aren't designed to mitigate the situation. As a result the probabilty of things going wrong is reasonably high. I'd say the bottom line is somewhere on the spectrum between poor design and negligence. Given it's a consumer product and should not require training, I think it needs addressing indeed.

The best way to address it would be backing down from this high power draw and having more efficient GPUs. The alternative would be switching to a higher-voltage bus to increase the margins, if ATX truly wants to support 600-800w expansion boards.

(If you're curious, this is also how USB-Cs newer PD modes like the 240w one get away with such a small conductor: It's 48V.)

Thanks for the explanations and sorry about my faux-pas (duh). English isn't my 1st language and I thought that "current" is the word for (German) "Spannung" but "Spannung" is actually "voltage".

I'm aware of the P = U * I formula which is Power = Voltage * Current/Amperage. I mistakenly assumed previously that the English word for 'U' is current when it is actually really simple as the correct English term is "voltage" and the unit is "Volt". Mea culpa .

Timmeh wrote on Feb 11, 2025, 19:29:

Coalfax wrote on Feb 11, 2025, 13:30:
Oh please AMD...Make the 9070 and 9070XT priced right and have useable features...

The 50 series is a joke by Nvidia to those that made them what they are today. They care not for us Gaming Plebs, only for the sweet AI cash they roll around in.

I think AMD is realizing they messed up. the 5000 series is barely faster than the 4000 series. AMD was clearly expecting Nvidia to have another large performance increase.

Fortunately, AMD only showcased a 9070 card. It seems like they can do a 9080 that is close to the 4090 and a 9090 card that is at least in the ballpark between the 4090 and 5090.

If they can do that and FSR 4 is as good as it looks like it is.... They can get a win this new generation. Especially if they price the cards hundreds lower and have avalability.

Keep dreaming .

The reason that RTX 5000 does not have a huge boost is because nVidia is using the (nearly) same 5nm manufacturing process as for RTX 4000.

AMD is doing the same with RDNA 4. They also delivered Zen 5 on a 5nm node in spite of 3nm availability.

Both companies are not interested in 3nm consumer products because they need all the 3nm wafers/capacities they can get for the AI/DC business. nVidia are making ~90% of their revenue from AI/DC. Gaming is officially chicken shit now.

AMD would be limited by the very same 5nm process limitations as nVidia so they can not just magically pull any cards out of the hat that all of a sudden beat everything (except the RTX 5090) that nVidia has to offer.

Mr. Tact wrote on Feb 11, 2025, 19:09:

RogueSix wrote on Feb 11, 2025, 18:03:
Faulty plugs happen. Just as faulty soldering (EVGA FTW!), faulty VRAM chips or faulty capacitors. When I said "flawlessly" then I meant within the usual margin of product deficiencies and I was also talking about the context at hand here.

If you go by the FUD that nVidia haters and AMD fanbois like to spread then they make it sound like every single card with that connector has burned to crumbs including the homes of the owners of the cards and their whole neighbahhoooodz.

Which is not quite the case in reality...

I am no AMD fanboy, despite having an AMD CPU. I haven't owned anything but an Nvidia video card for close to 20 years (possibly longer).

As I said, it is not a plug problem. It is how Nvidia is handling the power draw on the card. Are you claiming the removal of the shunt resistors and removal of dividing up the power draw is not a problem?

Now, do I know AMD is doing it better? Nope. That's why I said that would be a good question.

No, and I was not talking about you . You seemed to approach the whole thing on a factual basis contrary to some of the haters who claimed that 12VHPWR is a generally unsafe nVidia design.

It obviously isn't as there are millions of RTX 4000 cards with that connector out in the wild with marginal issues (probably real user error in quite a few cases and also poorly made 3rd party cables/adapters).

As I said, I think we need to wait and see if this is something RTX 5000 specific. The generalizations are just old and tired though. Contrary to what some people are claiming, this is not a connector that nVidia pushed alone but it is a PCI-SIG standard and was co-developed with AMD and others. It is working perfectly fine on millions of consumer and pro cards as we speak (including my own RTX 4090 since October 2022 and I love the clean one cable between PSU and card solution).

Anyway, the very low release supply actually makes me wonder if nVidia and their partners became aware of any such problems last minute and if they maybe already have a version 1.1 revision in production.

Timmeh wrote on Feb 11, 2025, 19:29:

Coalfax wrote on Feb 11, 2025, 13:30:
Oh please AMD...Make the 9070 and 9070XT priced right and have useable features...

The 50 series is a joke by Nvidia to those that made them what they are today. They care not for us Gaming Plebs, only for the sweet AI cash they roll around in.

I think AMD is realizing they messed up. the 5000 series is barely faster than the 4000 series. AMD was clearly expecting Nvidia to have another large performance increase.

Fortunately, AMD only showcased a 9070 card. It seems like they can do a 9080 that is close to the 4090 and a 9090 card that is at least in the ballpark between the 4090 and 5090.

If they can do that and FSR 4 is as good as it looks like it is.... They can get a win this new generation. Especially if they price the cards hundreds lower and have avalability.

They can have a big win in mainstream with 9070xt if they price it correctly. Making faster cards right now would be a waste of resources in my opinion as they are in a technological transition to UDNA next gen.

Sho wrote on Feb 11, 2025, 19:12:

RogueSix wrote on Feb 11, 2025, 16:46:
I'm no electrical engineer either but current = Volt. As the name 12VHPWR implies, this connector is made for 12V which isn't really "lots of current".

I'm really sorry, and I promise this isn't nitpicking either, but this is completely wrong. Voltage and current are not the same, and 12V is not a figure of current.

In fact, this problem would also be eliminated by moving to 24V or 48V, which have much lower resistive loss and wouldn't heat up to this degree.

The issue here is really simple. This is a 600W+ card that's pushing close to the spec envelope to the connector, and as a result has a very low safety margin. The electrics on the card aren't designed to mitigate the situation. As a result the probabilty of things going wrong is reasonably high. I'd say the bottom line is somewhere on the spectrum between poor design and negligence. Given it's a consumer product and should not require training, I think it needs addressing indeed.

The best way to address it would be backing down from this high power draw and having more efficient GPUs. The alternative would be switching to a higher-voltage bus to increase the margins, if ATX truly wants to support 600-800w expansion boards.

(If you're curious, this is also how USB-Cs newer PD modes like the 240w one get away with such a small conductor: It's 48V.)

Thank you for the detail.

Another solution: Increase power delivered by the PCIe slot. MSI had this announcement back in September, adding an extra 8-pin mobo power connector to juice the board and deliver "252W of extra power". I have not scene anyone actually test this with a 4090 or 5090, but it could protect the flaky 12VHPWR cable from spikes and sustained power draw at limit.

It's strange that MSI doesn't mention this anywhere on the product pages for these mobos. No detail in the manuals. And I've yet to see a YouTuber even mention it..................

Mr. Tact wrote on Feb 11, 2025, 19:09:
Now, do I know AMD is doing it better? Nope.

Yes you do, we all do.

You need only look at the extensive list of repairs to AMD GPU connectors.

What list you ask? Exactly.

Coalfax wrote on Feb 11, 2025, 13:30:
Oh please AMD...Make the 9070 and 9070XT priced right and have useable features...

The 50 series is a joke by Nvidia to those that made them what they are today. They care not for us Gaming Plebs, only for the sweet AI cash they roll around in.

I think AMD is realizing they messed up. the 5000 series is barely faster than the 4000 series. AMD was clearly expecting Nvidia to have another large performance increase.

Fortunately, AMD only showcased a 9070 card. It seems like they can do a 9080 that is close to the 4090 and a 9090 card that is at least in the ballpark between the 4090 and 5090.

If they can do that and FSR 4 is as good as it looks like it is.... They can get a win this new generation. Especially if they price the cards hundreds lower and have avalability.

RogueSix wrote on Feb 11, 2025, 16:46:
I'm no electrical engineer either but current = Volt. As the name 12VHPWR implies, this connector is made for 12V which isn't really "lots of current".

I'm really sorry, and I promise this isn't nitpicking either, but this is completely wrong. Voltage and current are not the same, and 12V is not a figure of current.

In fact, this problem would also be eliminated by moving to 24V or 48V, which have much lower resistive loss and wouldn't heat up to this degree.

The issue here is really simple. This is a 600W+ card that's pushing close to the spec envelope to the connector, and as a result has a very low safety margin. The electrics on the card aren't designed to mitigate the situation. As a result the probabilty of things going wrong is reasonably high. I'd say the bottom line is somewhere on the spectrum between poor design and negligence. Given it's a consumer product and should not require training, I think it needs addressing indeed.

The best way to address it would be backing down from this high power draw and having more efficient GPUs. The alternative would be switching to a higher-voltage bus to increase the margins, if ATX truly wants to support 600-800w expansion boards.

(If you're curious, this is also how USB-Cs newer PD modes like the 240w one get away with such a small conductor: It's 48V.)

RogueSix wrote on Feb 11, 2025, 18:03:
Faulty plugs happen. Just as faulty soldering (EVGA FTW!), faulty VRAM chips or faulty capacitors. When I said "flawlessly" then I meant within the usual margin of product deficiencies and I was also talking about the context at hand here.

If you go by the FUD that nVidia haters and AMD fanbois like to spread then they make it sound like every single card with that connector has burned to crumbs including the homes of the owners of the cards and their whole neighbahhoooodz.

Which is not quite the case in reality...

I am no AMD fanboy, despite having an AMD CPU. I haven't owned anything but an Nvidia video card for close to 20 years (possibly longer).

As I said, it is not a plug problem. It is how Nvidia is handling the power draw on the card. Are you claiming the removal of the shunt resistors and removal of dividing up the power draw is not a problem?

Now, do I know AMD is doing it better? Nope. That's why I said that would be a good question.

RogueSix wrote on Feb 11, 2025, 16:46:

Overon wrote on Feb 11, 2025, 16:25:
I'm no electrical engineer but isn't it stupid to put lots of current through thin wires?

I'm no electrical engineer either but current = Volt. As the name 12VHPWR implies, this connector is made for 12V which isn't really "lots of current".

The 12VHPWR spec is 9.5 A × 12 V × 6 pin = 684 W. The RTX 5090 has a max power draw of 600W so it is well within specs. The problem could be uneven power distribution (as Mr. Tact pointed out below, based on the reports and video from 8auer). There have been reports that some pins get fried by 20A which is more than double the max spec (9.5A).

Current is not the issue at hand. 12V is a joke. Static discharges when you walk on a carpet and touch something metallic can be in the KILO (thousands of) Volt range and you neither get fried nor die from it. It causes just a minor tickle.

We will have to wait and see. With the extremely low availability of 5090s, it is kind of hard to guesstimate how widespread the issue might be.

Indeed now I understand. So how much power does Nvidia save by only have 1 shunt resistor?

theglaze wrote on Feb 11, 2025, 17:46:

RogueSix wrote on Feb 11, 2025, 15:55:

Avus wrote on Feb 11, 2025, 14:27:
That plug is a sh!t design from day one and the most valuable company have too big of an ego to accept that.

Meanwhile, there are literally millions of RTX 4000 cards with this connector out in the wild that work flawlessly. The failure rate must be miniscule and is probably the same as the old clunky shitty garbage connector (yes, those also melt if not seated correctly).

Flawless huh?

Well, I RMA'd an RTX 40 series for a faulty 12VHPWR plug. Manufacturer didn't bat an eye at my claim, took 2 business days to replace the connector and ship it back. The tech did a (seemingly) flawless repair, only a different orientation of the backplate screws reveal it was even worked on. Besides the shiny new black plug.

A smooth process for a standard repair. As if 24 months of repetition has made it so..

Faulty plugs happen. Just as faulty soldering (EVGA FTW!), faulty VRAM chips or faulty capacitors. When I said "flawlessly" then I meant within the usual margin of product deficiencies and I was also talking about the context at hand here.

If you go by the FUD that nVidia haters and AMD fanbois like to spread then they make it sound like every single card with that connector has burned to crumbs including the homes of the owners of the cards and their whole neighbahhoooodz.

Which is not quite the case in reality...