RTX 4000 series burning cables thread

[Chrispy_]

Apparently it wasn't an FE.

Oh shit, I just went back to all W1zzard's 4090 reviews and every single one of them uses the HPWR connector :\

450W is too much for a graphics card and this bodge of adapters is a solution to a problem that shouldn't exist. Perhaps 5 years from now 450W will be considered normal but right now, it's just an obscene amount of power draw for a single component. How exactly is that supposed to scale to a laptop, or datacenter, for example?

 

[The red spirit]

How exactly is that supposed to scale to a laptop, or datacenter, for example?

Data center card, H100, already has different arch and 700 watt TDP for 1 GPU. Something like 4090 isn't meant to be scaled to lappys, there's going to be heavily cut down model for portables.

 

[Totally]

I mean,

If you have the money to buy a 4090GTX, u must have the money to buy a new PSU to go with it.

saving on PSU is not the play

Wtf do even mean?

When you get a new set of tires for your car, do you buy a whole new engine along with it?

 

[awesomesauce]

You’re comparing a video card with tires.

if my new set of tire were bigger, i would double check if my car could actually fit them.

the 4090 is an monster of GPU, not the day to day tire changing…

and if you in this kind of market (2000$USD/3K CAD gpu) you should be able to get another recent PSU. I donnot see why

 

[Bomby569]

Then how about we give it some time to see it addressed hey, my money is on the adapters themselves, but time will tell all, and then we can safely critisise whoever is responsible for it happening in the first place.

The topic isn't mine, i am commenting the topic and this case. Should we just ignore it?

I guess there is a reason people don't open topics related to "i have no issues on random hardware xy". And there is a bit more then just one person on reddit, you have cable mod, you have nvidia themselves, you have buildzoid, etc

 

[The red spirit]

The topic isn't mine, i am commenting the topic and this case. Should we just ignore it?

I guess there is a reason people don't open topics related to "i have no issues on random hardware xy". And there is a bit more then just one person on reddit, you have cable mod, you have nvidia themselves, you have buildzoid, etc

To be fair Buildzoid didn't encounter any problems firsthand, so he just says what he reads online.

 

[Bomby569]

To be fair Buildzoid didn't encounter any problems firsthand, so he just says what he reads online.

he did criticise the metal thing holding the actual cable inside the plastic housing. At least was what i got from his video. It's not a test sure.

 

[wolf]

The topic isn't mine, i am commenting the topic and this case. Should we just ignore it?

From my perspective, your commentary has been far from helpful, so perhaps ignoring it is good, or maybe I should just ignore you.

 

[Chrispy_]

Data center card, H100, already has different arch and 700 watt TDP for 1 GPU. Something like 4090 isn't meant to be scaled to lappys, there's going to be heavily cut down model for portables.

The H100 isn't even the same product segment, of course that's not what I'm meaning! VDI is just one of a few use-cases that require datacenter GPUs based on the desktop graphics architecture. The other is GPU rendering farms. VDI in particular is about a 25 billion dollar industry, annually - the fasted growing datacenter market at the moment, and likely to hit 33 billion USD in 2025.

3090 = RTX A5000 and A5500, running at 230W.
4090 = RTX L5000 and L5500? I'm guessing they'll use L for Lovelace....

We're not talking about one or two cards and one or two racks of niche-use servers here, we're talking tens of millions of units.

 

[TheDeeGee]

Looks like AMD changed their mind.

https://overclock3d.net/news/gpu_di..._connector_with_their_radeon_rx_7000_series/1

I wonder how far into the production stage AIBs are for 4070s, 4060s and 4050s. Maybe they can still change the design, or would that be too late? After all they would have to alter the production line as well.

 

[The red spirit]

The H100 isn't even the same product segment, of course that's not what I'm meaning! VDI is just one of a few use-cases that require datacenter GPUs based on the desktop graphics architecture. The other is GPU rendering farms. VDI in particular is about a 25 billion dollar industry, annually - the fasted growing datacenter market at the moment, and likely to hit 33 billion USD in 2025.

3090 = RTX A5000 and A5500, running at 230W.
4090 = RTX L5000 and L5500? I'm guessing they'll use L for Lovelace....

We're not talking about one or two cards and one or two racks of niche-use servers here, we're talking tens of millions of units.

You specifically mentioned data center though and that's what Hopper is for. Formerly Quadro line-up was for workstation, mostly VFX, CAD, graphical design. Also 3090 equivalent workstation card was RTX A6000 too, not only RTX A5000, anyway, TDP is 300 watts, which has been standard for top tier workstation cards for basically more than decade, I think that even 3DLabs was probably already producing cards like that, but I may be wrong. Anyway, workstation segment has been rather liberal with power usage, but more concerned about perf/watt. Also why do you think servers would need graphics cards? At best you meant rendering farms, supercomputers. Anyway, I don't see any reason to be more concerned about those at all. I expect basically RTX 4090 with 300 watt TDP for those uses, BTW there's another Hopper variant, specifically with PCIe connector and 300 watts. I see no reason why you are so alarmed and BTW this thread isn't for crying about every single flaw of RTX 4090 card, but just about that connector burning or not burning.

 

[Fleurious]

I’m curious if anyone has done infrared views of the cables under load at varying degrees of bends.

My takeaway from all this though as we’re dealing with wattage/amperages such that cables/connections/etc really should factor in user error that is not uncommon in the consumer space. Even halo products like these arn’t safe from a poorly seated connecter or a case side panel pushing against them.

More curious about how well the 4090 undervolts. Can you shave 20% load off and only lose 2-3% performance?

 

[Chrispy_]

You specifically mentioned data center though and that's what Hopper is for. Formerly Quadro line-up was for workstation, mostly VFX, CAD, graphical design. Also 3090 equivalent workstation card was RTX A6000 too, not only RTX A5000, anyway, TDP is 300 watts, which has been standard for top tier workstation cards for basically more than decade, I think that even 3DLabs was probably already producing cards like that, but I may be wrong. Anyway, workstation segment has been rather liberal with power usage, but more concerned about perf/watt. Also why do you think servers would need graphics cards? At best you meant rendering farms, supercomputers. Anyway, I don't see any reason to be more concerned about those at all. I expect basically RTX 4090 with 300 watt TDP for those uses, BTW there's another Hopper variant, specifically with PCIe connector and 300 watts. I see no reason why you are so alarmed and BTW this thread isn't for crying about every single flaw of RTX 4090 card, but just about that connector burning or not burning.

VDI is datacenter. I've spent 1-2 days a week on it, at minumum, for 6+ years.
Hopper is also datacenter, but that's compute, not graphics, and not relevant to discussion of 4090 cards, power use, architecture.

 

[ThrashZone]

Hi,
Out of 113 posts how many actual people here have cables burnup using the adapter ?

 

[The red spirit]

More curious about how well the 4090 undervolts. Can you shave 20% load off and only lose 2-3% performance?

Absolutely yes, has been shown during first days after launch. Optimum tech, Tech YES City made videos about that, I think TPU even wrote an article about that. You can easily shave 100 watts and yes 4090 is cranked way too high, way past severe diminishing return point. Hell, even at half power it will deliver like 85% performance. Yes, stock power limit is really that stupid.

 

[ThrashZone]

Oh shit, I just went back to all W1zzard's 4090 reviews and every single one of them uses the HPWR connector :\

450W is too much for a graphics card and this bodge of adapters is a solution to a problem that shouldn't exist. Perhaps 5 years from now 450W will be considered normal but right now, it's just an obscene amount of power draw for a single component. How exactly is that supposed to scale to a laptop, or datacenter, for example?

Yep and he's already stated no issues at all.

 

[P4-630]

Oh shit, I just went back to all W1zzard's 4090 reviews and every single one of them uses the HPWR connector :\

And he mentioned he used the same adapter for all 4090 's he reviewed......

 

[solarmystic]

So new AMD(7000)cards are going to have the exact same problem as they using identical connector right?

Apparently not, according to Kyle Bennett from HardOCP's sources.

https://twitter.com/i/web/status/1584856217335517186

 

[karakarga]

There are first group 500 to 1500 Watt PSU's from year 2010 and later. Some of those having directly leaving cables from PSU's inner left hand, are directly soldered inside the PSU's mainboard. Those are better actually. But, the other type, PSU chassis to inner connector card ones, I mean the first generation semi-pluggable cable having PSU's, GENERALLY DOES NOT HAVE, MATCHING AMOUNT OF CABLES internally!

What I mean is, if there are lets say 3 x 6 pin and 3 x 8 pin cables for graphics.

There must be 3x3=9 set (+12 Volt) for 6 pin and 3x3=9 set (+12 Volt) = [9+9] 18 set of cables for (+12 Volt) and
There must be 3x3=9 set (-12 Volt) for 6 pin and 3x5=15 set (-12 Volt) = [9+12] 21 set of cables for (-12 Volt)

But, if you count the cable amount inside the PSU you may see, (as an example) there may be only 12 set for (+12 Volt) instead of 18 and only 15 set for (-12 Volt) instead of 21!

--> So, if that is the case, do not use such old PSU's from that time! That might be the reason for heat, causing connector burns.... <--

Like old time Thermaltake Toughpower 500 to 1500 Watt, for example 1200A W0133RE or W0133RU and as well as Thermaltake Toughpower 575 Watt to 1475 Watt XT series. I personally count the cables inside many of them, the cable amount does not match....

Note: There are of course more pluggable 6 pin connector cables for 4 pin Molex and 5 pin power connectors, but again in sum, the cables count are not matching internally!

The new PSU's have directly soldered plug connector card units internally, they have full contact for the power draw!

 

[maxfly]

AMD chose not to use this world crushing, mind exploding connector.

There's your solution girls.

 

[N/A]

The solution is to use 16 - 24 pin microfit, but again early adopters pave the way.

 

[Tomgang]

In addition to my earlier post. Here are some pictures of how I have my adapter bent.

Also about connecting psu to none atx 3.0 standard like my own. Don't connect it with only two 8 pins that has a doublet. Meaning 2 x 8 pins on 1 cables. That means you will be pulling 300 watt from 1 cable designed for 150 watt at max. That risk overloaded cables and the rails in the psu it self. Psu goes puff at worst case scenario. All ways connect with 3 or 4 separate cables for each 8 pin to depending on what card you have. I'm sure most people on TPU knows that. But yet again some might not think about it and that could potentially risk overload psu power rails. 4090 can definitely pull at least 550 watt at peak. I have seen my own card do it with max power target and overclock. But never very close to the 600 watt limit. But that is stil enough to pull way more than 150 watt per cable if only 2 are connected.

Here are my pictures. As you can see it litterly tough sidedoor window. I'm trying to avoid as little bent on adapter as possible.

 

[N/A]

One thing manufacturers could have included is an OTP over temperature protection on the terminals of the card. Obviously it allowed for a melting point to be reached which is stupid. Especially on a 1799 video card is a must have.

 

[Totally]

Missed the whole point of the analogy, if it's adequate, and in good condition why does it need to be replaced.

the 4090 is an monster of GPU, not the day to day tire changing…

And the people buying don't know that? There aren't many options north of the1-1.5kw units out there that are seeing the cables burnt. Then to bring up PSU req's, NVIDIA's minimum recommendation is 850w and those are everywhere. So you're entire argument is just an incorrect assumption and a strawman.

 

[ir_cow]

The solution is to use 16 - 24 pin microfit, but again early adopters pave the way.View attachment 267119

So back to the 30 series connector?