News Nvidia RTX 5090's 16-pin power connector hits 150C in reviewer's thermal camera shots

[Admin]

Some thermal imagery shared on Twitter/X underlines how toasty-hot power connectors servicing Nvidia GeForce RTX 5090 graphics cards can get.

Nvidia RTX 5090's 16-pin power connector hits 150C in reviewer's thermal camera shots : Read more

 

[usertests]

If you're going to force everyone to adopt a new connector, how about making something that isn't the same Molex design from decades ago, but worse?

These things need to burn right off the market.

 

[spiketheaardvark]

It's there a reason we have decided to use lots of little wires to supply the most power hungry part of a computer instead of two big wires with a nice secure connection?

 

[ezst036]

Can we see thermal imaging from a 3090?

 

[tom2tec]

Classism is getting bad when our top corporations are this corrupt and incompetent. Where is the safety certification happening? Why were these ever approved? The public deserves answers.

 

[VizzieTheViz]

Reading this a short time after the article where MSI and NVidia say this connector is totally cool and not a fire hazard at all is pretty funny.

 

[Alex/AT]

It's there a reason we have decided to use lots of little wires to supply the most power hungry part of a computer instead of two big wires with a nice secure connection?

There actually is a reason. It's not only about wire size, it's also about component size and board interconnect volume (size). It's not enough to put single high-loaded wire to the board, you also need to carry this power further through the board, and components must be able to sustain the flow. Thus, multiple small wires are easier to maintain than just two large ones. And no, splitting it right on board is not going to be easy/funny as well (look at how stuff is split inside 1000W PSU and what size it all is...).

Considering this, just 4 x normal 8-pin PCI power connectors would've been a better job. They have enough sturdiness to spare and they are extremely time proven. Also loss of a single cable or slight misbalance won't kill them with wire count this high and GPU would have time to just power-throttle instead of burning up.

I suspect the use of 'brand new hot burning' connector also stems from a requirement it all must be from a single balanced bus (phase), and doing 4x8pin may lead to user errors of connecting these from different PSU buses (high power PSUs tend to have more than one circuit / bus) which will in turn require careful board design to not surge between the buses on the board itself. But this is totally doable given even GPU is totally not uniform inside.

Still, VHPWR must die either way, and that's the bottom line. It's almost an epitome of bad design. Burn baby burn.

 

[Heat_Fan89]

Some thermal imagery shared on Twitter/X underlines how toasty-hot power connectors servicing Nvidia GeForce RTX 5090 graphics cards can get.

Nvidia RTX 5090's 16-pin power connector hits 150C in reviewer's thermal camera shots : Read more

Get out the Bat fire extinguisher !

 

[Notton]

It's there a reason we have decided to use lots of little wires to supply the most power hungry part of a computer instead of two big wires with a nice secure connection?

There kind of was, back when PSUs had multiple 12V rails and the 8-pin EPS and PCIe were on separate ones.
There's supposed to be OCP (Over Current Protection) on each 12V rail, which was supposed to prevent cable fires, along with prevent shorts from destroying other components in the system.

Multiple 12V rail PSUs were the hot new thing.... I think in 2000~2005? because CPUs and GPUs were starting to pull enough power to require an extra 4-pin EPS and 6-pin PCIe. Mysteriously, the pin-out for 8-pin EPS and 8-pin PCIe are different (foreshadowing).

But then somewhere along the line Nvidia/PCISIG decided to throw all of that out the window and we're back to a single 12V rail that can pump 600W over 2 cables and a high resistance connector.

Now, the real kicker is, Nvidia/PCISIG didn't have to use this idiotic 16-pin fire connector.
The XT90, which can handle 90A, designed for Lithium batteries, in RCs existed since at least 2015. IDK the exact date, but the fire hazard 16-pin connector was introduced in 2022. Even if Nvidia was prototyping the 16-pin in 2019/2020, the XT90 connector predated it by at least 5 years.

 

[Alex/AT]

Mysteriously, the pin-out for 8-pin EPS and 8-pin PCIe are different (foreshadowing).

Same stuff I presume. Multiple bus (rails), so spikes from high-power peripheral rail can't hurt high frequencies-sensitive board/CPU rail. Thus these were deliberately made different to disallow mixup.

Modern 1000W+ PSUs also are easily encountered multiple-railed. Some offer to mix rails, but efficiency depletes and surge/noise level soars.

 

[TJ Hooker]

Multiple 12V rail PSUs were the hot new thing.... I think in 2000~2005? because CPUs and GPUs were starting to pull enough power to require an extra 4-pin EPS and 6-pin PCIe. Mysteriously, the pin-out for 8-pin EPS and 8-pin PCIe are different (foreshadowing).

But then somewhere along the line Nvidia/PCISIG decided to throw all of that out the window and we're back to a single 12V rail that can pump 600W over 2 cables and a high resistance connector.

The industry was transitioning back to single rail years before 12VHPWR came on the scene.

Modern 1000W+ PSUs also are easily encountered multiple-railed.

Do you have some examples of modern PSUs with multiple 12V rails? Because AFAIK single rail has been more common for a decade or more, even for high power PSUs.

 

[Shiznizzle]

Talk about built in obsolescence on steroids.

150 is good enough to cook eggs with. Some items i cook in my kitchen, cook in the oven at that temp. Who needs an oven to cook with when you can place your sunday roast on your GPU. The temp only gets turned up at the end to make sure middle reaches 160.

Nvdia, i think, this series will be avoided like the plague. I would not buy one. I wont buy one.

What the heck were they thinking when they designed this? Seriously. A multi-billion dollar corporation greenlit this garbage? Who da heck approved this mess?

If the public sees 150+ then you can be damn sure they did as well......or maybe not actually. That is a real possibility. They cant be playing dumb. No way.

 

[tootercomputer]

At what point do these connectors become a genuine fire hazard?

 

[John Kiser]

The industry was transitioning back to single rail years before 12VHPWR came on the scene.

Do you have some examples of modern PSUs with multiple 12V rails? Because AFAIK single rail has been more common for a decade or more, even for high power PSUs.

Single rail was also sought after when those companies started to do multiple 12v rails as often times the individual rails had lower amperage and could t keep up with certain compi enta if they got power hungry enough.

 

[John Kiser]

There kind of was, back when PSUs had multiple 12V rails and the 8-pin EPS and PCIe were on separate ones.
There's supposed to be OCP (Over Current Protection) on each 12V rail, which was supposed to prevent cable fires, along with prevent shorts from destroying other components in the system.

Multiple 12V rail PSUs were the hot new thing.... I think in 2000~2005? because CPUs and GPUs were starting to pull enough power to require an extra 4-pin EPS and 6-pin PCIe. Mysteriously, the pin-out for 8-pin EPS and 8-pin PCIe are different (foreshadowing).

But then somewhere along the line Nvidia/PCISIG decided to throw all of that out the window and we're back to a single 12V rail that can pump 600W over 2 cables and a high resistance connector.

Now, the real kicker is, Nvidia/PCISIG didn't have to use this idiotic 16-pin fire connector.
The XT90, which can handle 90A, designed for Lithium batteries, in RCs existed since at least 2015. IDK the exact date, but the fire hazard 16-pin connector was introduced in 2022. Even if Nvidia was prototyping the 16-pin in 2019/2020, the XT90 connector predated it by at least 5 years.

Okay so some things to get into. Nvidia while using the power connector is not the one responsible for putting them on AIB partner boards.. It should not have been finalized as a standard if it wasn't safe and also power supply makers need to be on the hook here at least a little bit for providing shorty wires and casing as far as what they can handle

 

[umeng2002_2]

lol

 

[beyondlogic]

its absolutely ludicrous that these connectors get to 150c is a absolute joke

 

[naryfa]

I have just the right song for this occasion:

View: https://www.youtube.com/watch?v=wmin5WkOuPw

 

[alceryes]

There actually is a reason. It's not only about wire size, it's also about component size and board interconnect volume (size). It's not enough to put single high-loaded wire to the board, you also need to carry this power further through the board, and components must be able to sustain the flow. Thus, multiple small wires are easier to maintain than just two large ones. And no, splitting it right on board is not going to be easy/funny as well (look at how stuff is split inside 1000W PSU and what size it all is...).

The 12VHPWR (and 12V-2X6) terminate into just one 12V connection on the GPU and inside the PSU. Neither the GPU nor the PSU see individual wires - it's just one 12V connection.

 

[teeejay94]

Man if people only realized what 8 pin and molex can already handle.

 

[Notton]

Okay so some things to get into. Nvidia while using the power connector is not the one responsible for putting them on AIB partner boards.. It should not have been finalized as a standard if it wasn't safe and also power supply makers need to be on the hook here at least a little bit for providing shorty wires and casing as far as what they can handle

1. Nvidia has total top down control approach to AIB cards. As an AIB, you either put the 16-pin power connector on the card, or you don't get nvidia chips.
2. Yes, the 16-pin power connector shouldn't have been a spec to begin with. You can blame PCI-SIG for approving that.
3. 🙄Please go educate yourself on entry level electricity. Hint: The cables are fine, they don't add much resistance. The connector is what heats up and melts because that's where the majority of resistance is. The various PSU makers are only building according to spec.

 

[qwertymac93]

It's there a reason we have decided to use lots of little wires to supply the most power hungry part of a computer instead of two big wires with a nice secure connection?

1. Cable flexibility.
2. Ability to use a single wire gauge for all connectors in the PC.
3. Redundancy so a single failed wire/connection isn't catastrophic.

When the ATX spec was made, even the newer revisions, nobody expected "peripheral cards" using more power than an entire PC to become common.

Something better than the standard 8-pin power connector was needed because otherwise we'd be looking at cards with 4 or even 5 power connectors hanging off the side. The sense pins were a good idea put into the new 16-pin connector I guess but the spec overall seems rushed and not fully baked.

In reality, the connector should have been limited to 450w and the spec should've required that it be treated as three sets of wires internally, as it was on the 3090/3090ti. We'd still be looking at 5090s with two power connectors, but at least they wouldn't be bursting into flames with "900w" of connector spec available to them.

Even better, a 12 pin connector based on the EPS12V connector with some sense pins. "Only" 336w per connector, but they'd be the larger, more robust pins that we know are tried and true and these ultra high end cards could still get away with two connectors. Also, fewer pins would mean easier insertion force and thus, less of a chance for "user error".

 

[YSCCC]

Reading this a short time after the article where MSI and NVidia say this connector is totally cool and not a fire hazard at all is pretty funny.

No, of course it isn't a fire hazard, why would you think that is? you... you are so utterly wrong for 2 reasons:

1) Orange Fire, the lowest temp of those, are 1100C, and 150C is what? it only boils water, so it is not a fire hazard, it's a water boiling thing

2) It's not a hazard, it's a feature 😉

so, it should be called a water boiling feature

/s

 

[closs.sebastien]

I don't understand why a simple connector seems so difficult to get it right.
A standard molex, or standard thing from sata, or even electrical/industrial world, or even from RC cars . can all withstand much more power without any problem.
Look at the DEAN connector of RC-Cars - or tamiya connector.
Many DC connectors exist, without any problem. https://en.wikipedia.org/wiki/DC_connector
Maybe should nvidia'co hire a simple electrical-technician (electric and not electronic)

 

[mtrantalainen]

There actually is a reason. It's not only about wire size, it's also about component size and board interconnect volume (size). It's not enough to put single high-loaded wire to the board, you also need to carry this power further through the board, and components must be able to sustain the flow. Thus, multiple small wires are easier to maintain than just two large ones. And no, splitting it right on board is not going to be easy/funny as well (look at how stuff is split inside 1000W PSU and what size it all is...).

You do realize that the Nvidia RTX 5090 reference design connects all those tiny wires on the PCB immediately after the connector, right?

You could easily replace the connector with a pair of thick wires soldered on the PCB without any additional changes to the design.

I'm looking forward somebody doing exactly this, using XT-90 connector, and then sharing images from thermal camera demonstrating that this would be far superior solution.