5090, nothing learned from the 4090 problems...

[TZeer]

Seems the 5090 is even worse than the 4090 when it comes to potentially melting cables.

130-150 degrees celcius on the actual connection. Uneven powerdistribution through the cable. Does not look good.

 

 

[Nightdare]

600W cable, GPU's pulling 630 watts consistently,... what could possibly go wrong?

[LucShep]

58 minutes ago, TZeer said:

Seems the 5090 is even worse than the 4090 when it comes to potentially melting cables.

130-150 degrees celcius on the actual connection. Uneven powerdistribution through the cable. Does not look good.

 

Yep.
And he already did a video on the RTX4090 before, explaining issues.*

If the 5090 is now pushing upto 25% more power than the previous 4090, then of course it's now a problem of "when", not "if".

To undervolt the 5090 is an absolute "must do", even more than before.

To put it simply, this connector should have never been used. And if so insisting on it, it should have been two of them in the 5090 and 4090.

* If you're short on time or patience, skip to 13:27 time of video:

 

Edited February 11 by LucShep

[Panzerlang]

Hopefully somebody (or a class-action) will sue the spivs.

[scommander2]

Questions:

1, will the DIY cable make different than Nvidia provided cable?  

2. will DCS draw that level of power from PS?

Thanks.

[SharpeXB]

The 5 series cards can also use the improved 12V-2x6 design. Definitely a worthwhile replacement if you’re going that route. I’m sure these YouTubers know that despite the clickbait titles
https://www.corsair.com/us/en/explorer/diy-builder/power-supply-units/what-power-cable-does-the-nvidia-geforce-rtx-5090-use/

41 minutes ago, Panzerlang said:

Hopefully somebody (or a class-action) will sue the spivs.

Already happened it was dismissed

https://www.classaction.org/news/nvidia-geforce-rtx-4090-melting-class-action-alleges-graphics-card-sold-with-defective-power-cable-plug-socket#:~:text=Filed: November 11%2C 2022 ◆,org's free weekly newsletter here.

 

Edited February 11 by SharpeXB

[TZeer]

In general:

1: No, the cable is similar. The cable is nothing high-tech. Problem in the video is an extremely uneven distribution of power through the different wires. Some of the 3rd party cards
has monitoring to check for these problems. But Nvidia founders card has not. But regardless, it's a <profanity>ty design and extremely poor QA from Nvidia's side.

2: It depends how much load there is on the GPU.

[scommander2]

3 hours ago, TZeer said:

2: It depends how much load there is on the GPU.

Thanks for the feedback.... I totally agree that objects are on terrains will make more loads and objects are like trees, AI planes... etc.

I have to salute to the owned 50 series pioneers who share their issues so that Nvidia and other venders can improve the products. 

Edited February 11 by scommander2

[EightyDuce]

Part of the issue, in that particular case, seems to stem from nvidia not actually utilizing any of the monitoring of the sense pins to do proper load balancing.

Cable/connector spec is rated for 660w by Molex (600w certified by Nvidia), 5090 stock TDP is 575w. Real tight margins... Basically, 5090 should have two connectors with proper load balancing. 

[Aapje]

Nvidia first makes a design with almost no margin of error, and then they don't even build in some security so that you can't have all the power going over one or two wires:

 

 

It's an absolute disgrace that this is what you get for (at least) $2k.

[TheBiggerBass]

This excessive power consumption - and price tag - are the main reasons why I don't want a 5090 or even 4090 in my system. My 4070 ti super already sinks about 300W. That has to be enough.

[BitMaster]

HOW STUPID must Nvidia be to NOT BE ABLE to produce a reliable 600w cable.

No rocket science, no nuclear tricks, no 10k kpa, PLAIN NOTHING that rests on their side of the scale. 

Almost ANY household appliance I have has 600+ watts, just to give a picture guys.

No, Nvidia refuses to learn the lesson.

 

This is so sad.  and I will not suggest, sell, install, or fix any of those cards. Nada

[Panzerlang]

7 hours ago, SharpeXB said:

Already happened it was dismissed

https://www.classaction.org/news/nvidia-geforce-rtx-4090-melting-class-action-alleges-graphics-card-sold-with-defective-power-cable-plug-socket#:~:text=Filed: November 11%2C 2022 ◆,org's free weekly newsletter here.

 

Quite clearly the judge took a brown envelope from NVidia. There was no legitimate defence available for NVidia and absolutely no possible way for that case to be legitimately dismissed. Yet another egregious example of corrupt corporations leveraging a corrupt judiciary.

[SharpeXB]

12 minutes ago, Panzerlang said:

Quite clearly the judge took a brown envelope from NVidia. There was no legitimate defence available for NVidia and absolutely no possible way for that case to be legitimately dismissed. Yet another egregious example of corrupt corporations leveraging a corrupt judiciary.

They admit guilt by putting a better connector on their next card I got to experience this one firsthand and now I’m the owner of the world’s most expensive drink coaster. 

[Pilotasso]

yeah I had issues with the 4090 (ASUS STRIX). The 12VHPWR cable adapter that came in the box caused me stability issues. The card showed a red LED at the power plug indicating something was wrong but it still was capable of runing becnhmarks even. Then I bought a cablemod with 12VHPWR PCI-e with 4 connections to the PSU. LED still showed red and problems continued.

Then I bought an ATX 3XPCI-E->to->12VHPWR cable from the PSU manufacurer (wasnt initialy available), LED came off and that has been rock solid ever since. The cables dont heat up.

Just dont use 3rd party cables. If you dont have an adapter from the PSU brand just save yourself trouble and buy a new PSU 1000W+ with one dedicated 12V plug and cable.

Edited February 11 by Pilotasso

[LucShep]

Honestly, the more I read about this, the more I find this (far) worse than what happened with Intel Raptor Lake CPUs. 
 

8 hours ago, EightyDuce said:

Part of the issue, in that particular case, seems to stem from nvidia not actually utilizing any of the monitoring of the sense pins to do proper load balancing.

Cable/connector spec is rated for 660w by Molex (600w certified by Nvidia), 5090 stock TDP is 575w. Real tight margins... Basically, 5090 should have two connectors with proper load balancing. 

Exactly.
No security monitoring and too little margin of error.

The power port on the 5090 overheats and burns because the current isn’t evenly distributed across the pins.
In addition, it should have had at least a second connector (not just one).

As the Der8auer tests show, one of the 12-pin connector wires draw 23 amps (over double of what is suppposed to run?!?) then causing temperatures to spike to 150ºC. 
That's crazy.
A connector heating up to 70ºC, while the PSU gets extremely hot at 150ºC, all in just a few minutes of benchmark testing, is perhaps all that needs to be seen. 

And now think about this - if having the mentioned headroom of 5% (Nvidia 600W rating) or 15% (Molex 660W rating) is already too little on itself, then one has to wonder about the possible transient spikes, possibly going over 750W(?) in that puny cable and connector on each end, all prone to manufacturing tolerances....

What I wonder now is, seeing how this can get dangerous real quick, how could this have passed safety tests for consumer market etc? 
 

7 hours ago, Aapje said:

Nvidia first makes a design with almost no margin of error, and then they don't even build in some security so that you can't have all the power going over one or two wires:

 

 

It's an absolute disgrace that this is what you get for (at least) $2k.

Always good videos from Buildzoid. 
 

Edited February 12 by LucShep

[Dogmanbird]

It's been 6 months of disappointment. msfs2024, Assetto Corsa Evo and now RTX5090

They usually come in threes so hopefully the next thing to come out will be a winner

Maybe AMD?

 

 

[Aapje]

11 hours ago, Panzerlang said:

Quite clearly the judge took a brown envelope from NVidia. There was no legitimate defence available for NVidia and absolutely no possible way for that case to be legitimately dismissed. Yet another egregious example of corrupt corporations leveraging a corrupt judiciary.

The reporting suggests that Nvidia simply paid off the person who filed the suit and they withdrew the complaint. There was never a ruling.

[Aapje]

17 hours ago, LucShep said:

What I wonder now is, seeing how this can get dangerous real quick, how could this have passed safety tests for consumer market etc? 

In the EU, the CE-marking is merely a legal construct. Putting it on the product is considered to be a promise by the manufacturer that they ensured that their product complies with the EU standards. So if they put the CE-marking on the product without actually doing that, the courts can conclude that it was not a mere oversight, but a willful act of not following the law. But there is no testing required by an independent/government body.

Testing it would be hard anyway, since the Low Voltage Directive just uses generic language stating that the product should be safe to use and connect, and that the manufacturer should recall or fix the device if this turns out not to be the case (despite a solid effort to make it safe). Presumably, the courts would create jurisprudence, or it already exists, on what is considered to be safe enough, based on expert testimony or the assessment by national agencies, and when a recall is warranted.

In the US, the laws seem more centered around empowering agencies to make and enforce rulings, but GPUs are probably not on the radar of any safety agency right now.

So in practice, we probably either need a sufficiently big scandal with people dying for this to get on the radar of the agencies, or people need to sue themselves.

Edited February 12 by Aapje

[Panzerlang]

I'm going to guess that unless NVidia publicly acknowledges the flaw and just as publicly shows the fix, the sales of the 50 series will take a very serious hit. However, I wouldn't be at all surprised to learn it's a deliberate ploy to get out of the gaming market with less damage to their rep than simply saying "screw all you gamers", so they can go all-in with chips for AI only.
It would explain the lack of cards at launch too, they wanted to minimize the chances of actually causing death by fire before they got their manufactured excuse to bail.

Edited February 12 by Panzerlang

[LucShep]

 

 

2 hours ago, Aapje said:

12 hours ago, LucShep said:

What I wonder now is, seeing how this can get dangerous real quick, how could this have passed safety tests for consumer market etc? 

In the EU, the CE-marking is merely a legal construct. Putting it on the product is considered to be a promise by the manufacturer that they ensured that their product complies with the EU standards. So if they put the CE-marking on the product without actually doing that, the courts can conclude that it was not a mere oversight, but a willful act of not following the law. But there is no testing required by an independent/government body.

Testing it would be hard anyway, since the Low Voltage Directive just uses generic language stating that the product should be safe to use and connect, and that the manufacturer should recall or fix the device if this turns out not to be the case (despite a solid effort to make it safe). Presumably, the courts would create jurisprudence, or it already exists, on what is considered to be safe enough, based on expert testimony or the assessment by national agencies, and when a recall is warranted.

In the US, the laws seem more centered around empowered agencies to make and enforce rulings, but GPUs are probably not on the radar of any safety agency right now.

So in practice, we probably either need a sufficiently big scandal with people dying for this to get on the radar of the agencies, or people need to sue themselves.

Woooaaa  ...I had no idea it was that lenient! (and thanks for the explanation)

I suppose that, yes, it'll have to get far worse before it gets any better.  

The next coming months will be revealing, I guess.
 

16 hours ago, BitMaster said:

HOW STUPID must Nvidia be to NOT BE ABLE to produce a reliable 600w cable.

No rocket science, no nuclear tricks, no 10k kpa, PLAIN NOTHING that rests on their side of the scale. 

Almost ANY household appliance I have has 600+ watts, just to give a picture guys.

No, Nvidia refuses to learn the lesson.

 

This is so sad.  and I will not suggest, sell, install, or fix any of those cards. Nada

+1. Ditto
 

58 minutes ago, Panzerlang said:

I'm going to guess that unless NVidia publicly acknowledges the flaw and just as publicly shows the fix, the sales of the 50 series will take a very serious hit. However, I wouldn't be at all surprised to learn it's a deliberate ploy to get out of the gaming market with less damage to their rep than simply saying "screw all you gamers", so they can go all-in with chips for AI only.
It would explain the lack of cards at launch too, they wanted to minimize the chances of actually causing death by fire before they got their manufactured excuse to bail.

It'd be a shame to be without future GPUs from the leading hardware/software manufacturer in the area.

But then, if this is really their modus operandi, releasing products on the verge of price gouging while being potentially faulty (and dangerous), then I honestly think we'd be better without them (good riddance).

There'd still be AMD and Intel (and possibily others who'd venture, like in past) picking where it'd be left at.  

 

Edited February 12 by LucShep

[Nightdare]

1 hour ago, LucShep said:

It'd be a shame to be without future GPUs from the leading hardware/software manufacturer in the area.

 

On the other hand, it would force game coders to start programming and optimizing their games more efficiently

Because this would be an admit to defeat to increase the performance of hardware for fidelity

[scommander2]

1 hour ago, LucShep said:

it'll have to get far worse before it gets any better.

Yup, everything is going to display on table.

[SharpeXB]

It would sure seem to be the best move for anyone getting this card to use the 12V-2x6 connector. Not the 12VHPWR

https://www.corsair.com/ww/en/explorer/diy-builder/power-supply-units/evolving-standards-12vhpwr-and-12v-2x6/

Edited February 12 by SharpeXB

[LucShep]

11 hours ago, SharpeXB said:

It would sure seem to be the best move for anyone getting this card to use the 12V-2x6 connector. Not the 12VHPWR

https://www.corsair.com/ww/en/explorer/diy-builder/power-supply-units/evolving-standards-12vhpwr-and-12v-2x6/

There's still no security monitoring.
The current is still not evenly distributed across the pins.
There's still too little margin of error with one single connector.

The problem remains, i.e. the difference with the new cable exhists but issues are only better disguised.

Edited February 13 by LucShep