NVIDIA RTX 50 SERIES - Technical/General Discussion

[gbgt2]

I've been using the 4 x 8-Pin adapter with my 4090 for over 2 years with no issues. Based on what I have seen I would be tempted to do the same thing with a 5090 if I had one since it seems that allows the PSU to balance the current more effectively. Downside is obviously the mess of cables ..

Edit: Actually I think this means you are limited to 450W? So big downside lol

 

[andybird123]

I've been using the 4 x 8-Pin adapter with my 4090 for over 2 years with no issues. Based on what I have seen I would be tempted to do the same thing with a 5090 if I had one since it seems that allows the PSU to balance the current more effectively. Downside is obviously the mess of cables ..

Edit: Actually I think this means you are limited to 450W? So big downside lol

No, only if you don't have all 4 of the connectors connected

 

[andybird123]

Derbauer has a video up of 5090 FE power draw, it's not drawing power evenly on over all 6 cables meaning one or two of them heat up significantly more than others. At the moment it looks like only the FE has this issue.

 

[Deathjester]

Derbauer has a video up of 5090 FE power draw, it's not drawing power evenly on over all 6 cables meaning one or two of them heat up significantly more than others. At the moment it looks like only the FE has this issue.

However given how the power distribution works on the 5090 FE (basically a big blob of 12V power through a single shunt resistor, but delivered to the GPU over 6 wires) there's no way for the FE to "decide" to draw that over one wire more than any other, the only real reason this can be is there is a significant resistance change (0.5 ohms to 6 ohms) across the 6 connections.

Either this is on the board, or it's a cable / connector issue. I would suggest the latter.

 

[Soulja]

For the people who a have the melting issues, whould they be able to get replacements for their gpu/psu or would they say its user error?

 

[Sinbad2000]

Not sure what you mean when you say "bus" as those are normally used in communications, do you mean busbar?

Also while there is some resistance drop when a conductor heats up you're talking about pretty small changes, maybe the 10% 'ish range by going from room temp to 100°C, the same sort of range you'd see from using a 6' wire instead of a 3' wire.

Yes sorry busbar. But even a 10% increase in resistance will mean a non-trivial drop in current compared to the other conductors all other things being equal.

 

[andybird123]

However given how the power distribution works on the 5090 FE (basically a big blob of 12V power through a single shunt resistor, but delivered to the GPU over 6 wires) there's no way for the FE to "decide" to draw that over one wire more than any other, the only real reason this can be is there is a significant resistance change (0.5 ohms to 6 ohms) across the 6 connections.

Either this is on the board, or it's a cable / connector issue. I would suggest the latter.

If it was a cable issue it would be repeatable with different cards, but he says it only happens on the FE card he has.

 

[StarShock]

Worth noting, it can work fine, and if its working correctly there's no issue. The problem comes in when there's a failure somewhere, as it seems instead of failing and stopping working it fails and catches fire.

 

[benjii]

For the people who a have the melting issues, whould they be able to get replacements for their gpu/psu or would they say its user error?

I was also wondering that. If both the PSU and GPU die, would one party replace both or would you have to contact both parties individually. I guess it depends on if blame is assigned.

 

[Murphy]

Yes sorry busbar.

They're used to distribute power to multiple connections, a bit like how adaptor plugs can make a single socket into multiple sockets. Do you have a link to where it was mentioned the FE has a bus on the card side because I'd be interested to know what they referring to.

But even a 10% increase in resistance will mean a non-trivial drop in current compared to the other conductors all other things being equal.

Yes but there's hundreds of things that can increase resistance by non-trivial amounts, if we were to expect the thing supply the power to account for all of them you'd have all sorts of problems.

Change the cable length = resistance change.
Change to solid core = resistance change.
Change from Alu to Cu = resistance change.
Change from poor to good quality Cu = resistance change
Change the gauge of the wire = resistance change.

There's good reasons for why the responsibility of how much power something can draw is not put on the thing supplying it, why we cut power to things that try to draw more power than the system can handle, etc, etc.

 

[Sinbad2000]

They're used to distribute power to multiple connections, a bit like how adaptor plugs can make a single socket into multiple sockets. Do you have a link to where it was mentioned the FE has a bus on the card side because I'd be interested to know what they referring to.

In DerBauer's video he mentions that all the HPWR connectors come into a single point.

, 16:30 in.

Yes but there's hundreds of things that can increase resistance by non-trivial amounts, if we were to expect the thing supply the power to account for all of them you'd have all sorts of problems.

Change the cable length = resistance change.
Change to solid core = resistance change.
Change from Alu to Cu = resistance change.
Change from poor to good quality Cu = resistance change
Change the gauge of the wire = resistance change.

There's good reasons for why the responsibility of how much power something can draw is not put on the thing supply it, why we cut power to things that try to draw more power than the system can handle, etc, etc.

Sorry I think you're missing the point of my post. I'm not saying the power supply should account for anything.
I'm saying that all other things being equal (ie, same potential (voltage) and same load at each conductor of a multi part cable), the amperage across all the conductors should balance out if the ends come to a common point - as cables carrying more power will get hotter and thus increase their resistance, dropping the current they'd carry). If the 5090FE 12V lines all come to a common point, the only way that 2 conductors can carry significantly higher current than the other 4 is if 1) there's something seriously wrong with the cable and 4 of the conductors have much higher resistance at the connectors) or 2) the PSU is somehow providing a lot more potential under load to those two conductors.

That being said, I'm sure it's absolutely possible for a PSU to regulate output current on each conductor. If the spec says 12 amps per conductor, then don't send more than 12 amps across it.

 

[Deathjester]

If it was a cable issue it would be repeatable with different cards, but he says it only happens on the FE card he has.

I know that Falcon Northwest published their own test with multiple FE cards and didn't see the same issues occuring.

However the AIB cards have a slightly different power setup to the FE, since the FE just rams everything through a single shunt resistor so you're going 1 rail to 1 rail, rather than performing any kind of power balancing. In contrast the ASUS Astral has 6 into 1 allowing you to set alerts if the utilization over one becomes too high.

This explains how the power configuration works between the different cards.

Yes, some of the fault lies with the lazy configuration of the FE power design, but since electricity follows the path of least resistance there is clearly an significant difference in resistance between one connection and the others; given that the difference is 23-11-8-5-3-2 (amps per wire) it would imply that something doesn't look like it's seated correctly and that's unlikely to lie at the PSU end since he is also using 2 connectors from the PSU to the GPU with 3 wires each.

 

[Murphy]

In DerBauer's video he mentions that all the HPWR connectors come into a single point.

That's not a bus or busbar. A bus or busbar splits a single input into multiple outputs, what he describes is the opposite of that (not sure there's an antonym for that).

Closets word i know is a series circuit but even then they're normally single source supplying multiple devices, like with fair-lights, not multiple sources supplying one device.

Sorry I think you're missing the point of my post. I'm not saying the power supply should account for anything.
I'm saying that all other things being equal (ie, same potential (voltage) and same load at each conductor of a multi part cable), the amperage across all the conductors should balance out - as cables carrying more power will get hotter and thus increase their resistance, dropping the current they'd carry)

I get what you're saying but the point I'm making is that electricity will follow the path of least resistance and when compared to all the other things than can change resistance a cable getting a bit hotter is barely a blip.

 

[Sinbad2000]

That's not a bus or busbar. A bus or busbar splits a single input into multiple outputs, what he describes is the opposite of that (not sure there's an antonym for that).

OK in my head a busbar basically converts (from a logical point of view) multiple contacts into one. Maybe that helps explain what I'm saying?

And the point you're making is exactly my point Path of least resistance... increase resistance even slightly, and electricity will take other paths, which results in the load balancing I'm talking about.

I think my language is just not 100% technically right which is causing confusion.

 

[Icyseanfitz]

Is there any indication of a better 5080 stock situation in the next month or two? Id actually buy one at or close to msrp but there's absolutely nothing, surely people aren't actually buying them en masse at the current ludicrous pricing. I've a 7900xtx on order because of it ‍

 

[Sc00p007]

Is there any indication of a better 5080 stock situation in the next month or two? Id actually buy one at or close to msrp but there's absolutely nothing, surely people aren't actually buying them en masse at the current ludicrous pricing. I've a 7900xtx on order because of it ‍

Fomo and hype is what the high end GPU scene has become.

I'd guess it will settle down in a month for 5080, scalper prices are already reducing and the drops are staying avalable for a little longer (although still above msrp)

 

[gpuerrilla]

Is there any indication of a better 5080 stock situation in the next month or two? Id actually buy one at or close to msrp but there's absolutely nothing, surely people aren't actually buying them en masse at the current ludicrous pricing. I've a 7900xtx on order because of it ‍

Sign up to one of the alert sites.

 

[Mysterae_]

In the Derbaur's video he touched upon this article on the Molex website:

https://www.molex.com/en-us/blog/gold-or-tin-vs-gold-and-tin

In my opinion, this is the crux of the problem. I'm sure it'll transpire that there is an issue with the coating of the pins of the GPU connector, PSU connector as well as any cable adaptors. Many of them don't state the type of coating used.

The problems may be a collection of:

- Poorly applied coatings
- Mixing of coatings (such as gold mated to tin as per the article above)
- Intermetallic compound formation - caused by heat and stress and can result in non-metal elements being formed
- Fretting corrosion - a type of metal fatigue that occurs when two surfaces rub together (vibration, high contact pressure, heating/cooling cycles) causing corrosion and wear

Something is causing an increase in resistance in some wires allowing the current to favour others, and the above are likely candidates.

This should have been mitigated in the design and ruled in the specification so that it was a non-issue but alas...

Choose your power connectors wisely!

 

[Monkeyboysteve]

Is there any indication of a better 5080 stock situation in the next month or two? Id actually buy one at or close to msrp but there's absolutely nothing, surely people aren't actually buying them en masse at the current ludicrous pricing. I've a 7900xtx on order because of it ‍

Doubt the stock situation will improve tbh. i used "hotstock" to get my 5080 at MSRP.
This is how I got my 3080 at MSRP years ago as well, and a PS5 during the shortages years ago at release.

 

[tommybhoy]

We asked a PSU expert which adapters or extensions are okay for the RTX 50-series. His answer: 'DO NOT BUY adapters or extenders, they can all be dangerous'

Spending $2,000 or more on a new GPU? Then get the right PSU and cables to go with it.

www.pcgamer.com

Short version:

Only use the supplied Cable shipped with your 50 series GPU.
Don't use any 3rd party cables:

Especially if it's a cable carried over from the 40 series as the 5090 draws up to 56 53 Amps through a (the newer)cable designed for 60 Amps.