Help with new PSU and 12VHPWR connector

[neurotix]

Hey all.

 

I got a brand new Superflower Leadex VII XG 1300w powering my kit currently. My use includes folding pretty much 24/7. I wanted to get opinions or make sure what we've done is safe.

 

 

 

Basically, we needed to kink the wire below the 12VHPWR connector to keep the connector itself straight, and to be able to put my side glass window on without it pushing down on the 12VHPWR at the actual connection which is bad.

 

Afaik on the GPU side it connects to two PCI-E 8-pin connectors and those are straight but in their advertising they said "safer design to prevent bend issue", implying that bending the PCI-E on the PSU side is okay. Wondering if bending the wire like we have on the GPU side is okay too.

 

Opinions or suggestions welcome, or even just a validation that what I'm doing is going to be safe long term for a relatively high demand scenario. Thanks in advance

 

Edited July 9 by neurotix

[Slaughtahouse]

* Disclaimer * I don’t own a PSU and/ or GPU with this connector.
 

Congrats on the purchase

 

The risk has been reviewed in depth by Gamers Nexus. The bend (within reason) has no impact. The source of issues stem from the male connector NOT being fully seated into the female connector.

 

As long as the bend isn’t putting strain on the connector or tension on the cable, in which the male connector is at risk of disconnecting, you’re fine.

 

Obviously if you put immense strain on the cable by bending it 180 degrees with a very tight radius, maybe the cable would fail at the bend. but you’re far from that situation. The reported issues that blew up in the media were isolated to the connector not being fully seated.

 

In other words, I think the bend is fine.

 

Source: https://youtu.be/ig2px7ofKhQ?si=mr3_CvLmU5rmfa1t

 

 

[neurotix]

24 minutes ago, Slaughtahouse said:

* Disclaimer * I don’t own a PSU and/ or GPU with this connector.
 

Congrats on the purchase

 

The risk has been reviewed in depth by Gamers Nexus. The bend (within reason) has no impact. The source of issues stem from the male connector NOT being fully seated into the female connector.

 

As long as the bend isn’t putting strain on the connector or tension on the cable, in which the male connector is at risk of disconnecting, you’re fine.

 

Obviously if you put immense strain on the cable by bending it 180 degrees with a very tight radius, maybe the cable would fail at the bend. but you’re far from that situation. The reported issues that blew up in the media were isolated to the connector not being fully seated.

 

In other words, I think the bend is fine.

 

Source: https://youtu.be/ig2px7ofKhQ?si=mr3_CvLmU5rmfa1t

 

 

Thank you so much @Slaughtahouse!!!

 

Will definitely double check to make sure it's seated fully.

[Slaughtahouse]

5 minutes ago, neurotix said:

Thank you so much @Slaughtahouse!!!

 

Will definitely double check to make sure it's seated fully.

 

If you’re in doubt, whilst the PSU and system is disconnected and without power, wiggle the cable. If it slides out of the female connector, you know it’s not seated correctly.

 

Highly recommend you take 30mns out of your day to watch that video linked above.

[neurotix]

7 minutes ago, Slaughtahouse said:

 

If you’re in doubt, whilst the PSU and system is disconnected and without power, wiggle the cable. If it slides out of the female connector, you know it’s not seated correctly.

 

Highly recommend you take 30mns out of your day to watch that video linked above.

 

Will do. I used to watch Gamer's Nexus and a lot of other YouTube channels, but haven't really since 2020. I'll watch it.

[J7SC_Orion]

...is there any way to re-route your 12VHPWR cable from the top ? I am quite paranoid about that cable (& often see more than 670 W peak on GPU) so no matter which cable and PSU combo I use for it (among several), I always have the 12VHPWR cable run straight down into the GPU connector. This latest PSU has a single-connector 12VHPWR at the PSU end as well, but did the same with my other combos.  

Those 12VHPWR connectors make me nervous no matter what re. the cross section and current it has to handle...

 

[Mr. Fox]

I have two 4090s and I have not had any issues. To be clear, the bend itself isn't the problem. The issues surface when the connector doesn't remain properly seated in the socket. As long as it remains fully seated and there are no weak connections inside of the socket due to poor construction and lousy contact  at the pins you should be fine.

 

The 12VHPWR cable is a fragile and ill-conceived engineering abortion that never should have seen the light of day. I would love to see multi-nation governmental involvement result in forcing NVIDIA to recall all video cards that use it and refund original purchase prices including taxes and shipping. They deserve to be severely harmed financially for having released this trash.

[Sir Beregond]

I guess if you were really nervous about it, are you really gaining much running 670W vs 450W on the 4090? Lack of power was never really the 4090's problem like it was with the 30-series. Hell 350W is basically within 5% of the performance of the card at 450W, so I have to imagine 670W is highly diminished returns. It's not 0, there are gains, but for an extra 220W seems minimal.

 

Now I'm not sure the difference in folding PPD vs gaming FPS, but something to think about if it really makes you worried.

 

All I know is the beefiest 12VHPWR cables are rated up to 600W, so you're probably maxing that out (and then some maybe) plus power draw from the PCIe slot. That's gotta be taxing the hell out of the PSU. If Nvidia sticks with this power connector, they really ought to put two of them on these higher wattage cards like the Kingpin or HOF cards did.

 

EDIT: Wrote this too early, didn't realize the 670W was @J7SC_Orion and peak power. What kind of normal power draw are you getting @neurotix? I thought you ran a 660W BIOS or something, but would be interested to hear.

 

 

Edited July 10 by Sir Beregond

[Slaughtahouse]

The cables that meet PCIe spec can support the 600w without issue. They should be 16 AWG and with that gauge can easily handle loads over 1000w. That's partly whythey can say "200% power excursions". 

 

Seasonic even has a nice image to help promote best practices. Superflower also has a similar image on their wesbiste. 

 

They're all recommending 3.5cm (1 3/8") distance between the tip of the connector and the start of the radius for the bend. There are no guidelines or max radius' specified on the bends. Either way, there haven't been any fires to my knowledge from the bends themselves. All the issues previously reported tie back to the issue reiterated by GN and Mr. Fox. Poor contact at the connector means path of least resistance, current goes through plastic, takes the load and melty melty starts. 

 

[neurotix]

2 minutes ago, Slaughtahouse said:

The cables that meet PCIe spec can support the 600w without issue. They should be 16 AWG and with that gauge can easily handle loads over 1000w. That's partly whythey can say "200% power excursions". 

 

Seasonic even has a nice image to help promote best practices. Superflower also has a similar image on their wesbiste. 

 

They're all recommending 3.5cm (1 3/8") distance between the tip of the connector and the start of the radius for the bend. There are no guidelines or max radius' specified on the bends. Either way, there haven't been any fires to my knowledge from the bends themselves. All the issues previously reported tie back to the issue reiterated by GN and Mr. Fox. Poor contact at the connector means path of least resistance, current goes through plastic, takes the load and melty melty starts. 

 

I'll have to measure it later then. Thanks again.

 

I will get to everyone else's posts later when I have time. Rep all around.

[neurotix]

20 hours ago, Slaughtahouse said:

 

If you’re in doubt, whilst the PSU and system is disconnected and without power, wiggle the cable. If it slides out of the female connector, you know it’s not seated correctly.

 

Highly recommend you take 30mns out of your day to watch that video linked above.

 

I did this yesterday and it was fine, cable is fully attached to the GPU. Now I'm paranoid about the 8-pins on the PSU side. Might have to check those and make sure they're in all the way.

 

8 hours ago, Mr. Fox said:

The 12VHPWR cable is a fragile and ill-conceived engineering abortion that never should have seen the light of day. I would love to see multi-nation governmental involvement result in forcing NVIDIA to recall all video cards that use it and refund original purchase prices including taxes and shipping. They deserve to be severely harmed financially for having released this trash.

 

Yeah, something definitely needs to be done about this. The government should get involved with how dangerous these things are and can be. People could end up burning their houses down at the worst or frying a way overpriced $1700 GPU at the least.

 

5 hours ago, Sir Beregond said:

 

EDIT: Wrote this too early, didn't realize the 670W was @J7SC_Orion and peak power. What kind of normal power draw are you getting @neurotix? I thought you ran a 660W BIOS or something, but would be interested to hear.

 

 

 

 

My stock MSI BIOS on the 2nd "performance" setting via the switch on it is 450w, with 108% power limit being the maximum in Windows, and 480w being the maximum in Linux and I presume Windows as well.

 

I did (notoriously) run the Galax 666w BIOS and folded on it 24/7 for like 4 months on that BIOS with a 2014-era Cooler Master V1000 (Seasonic) power supply. MSI includes a 12VHPWR dongle to three 8-pin PCI-E power. I was supposed to run three separate wires individually connected to the PSU to the connectors on that dongle, but I was stupid and lazy and daisy chained one of them. @bridgypoo pointed out the whole time that we should run another separate wire, but I was like "oh no my PSU is Seasonic and high quality and it'll be fine". These were NOT ATX 3.0 or whatever PCI-E 8 pins so their max wattage should be 150w, and 150w x3 is 450, just enough to meet the power limit of the card. I ended up frying cables and ports on the back of that PSU.

 

Anyway, I have a new Leadex 1300w unit since my Seasonic Vertex GX died, we got approved for RMA though. Its powering my folding just fine so far.

 

I do not use the 666w bios, but I do raise the power limit and overclock the card using an app in Linux. See picture:

 

 

Note the power draw in GreenWithEnvy. Folding takes 350w, to 400w max according to this. If I look at the Kill-a-Watt meter that just the box is plugged into, it's usually drawing 450-550 total. (And it's a 1300w psu) This new PSU is dual rail, the Seasonic Vertex we are RMAing was not. I don't fold on the CPU, just on the 4090.

 

Sorry that was so long.

 

1 hour ago, Slaughtahouse said:

The cables that meet PCIe spec can support the 600w without issue. They should be 16 AWG and with that gauge can easily handle loads over 1000w. That's partly whythey can say "200% power excursions". 

 

Seasonic even has a nice image to help promote best practices. Superflower also has a similar image on their wesbiste. 

 

They're all recommending 3.5cm (1 3/8") distance between the tip of the connector and the start of the radius for the bend. There are no guidelines or max radius' specified on the bends. Either way, there haven't been any fires to my knowledge from the bends themselves. All the issues previously reported tie back to the issue reiterated by GN and Mr. Fox. Poor contact at the connector means path of least resistance, current goes through plastic, takes the load and melty melty starts. 

 

 

I'm about to turn my rig off, take the glass off and measure. Thanks for this.

Edited July 10 by neurotix

[neurotix]

I measured the length of the cable before the bend and it was about 1 3/4" so according to those charts, it should be ok (I hope). I am shocked@piois not in this thread making fun of me I fried another one bro, a Seasonic Vertex GX 1200w that was only a year and 3 months old. Lol.

[Sir Beregond]

1 hour ago, neurotix said:

 

I did this yesterday and it was fine, cable is fully attached to the GPU. Now I'm paranoid about the 8-pins on the PSU side. Might have to check those and make sure they're in all the way.

 

 

Yeah, something definitely needs to be done about this. The government should get involved with how dangerous these things are and can be. People could end up burning their houses down at the worst or frying a way overpriced $1700 GPU at the least.

 

 

My stock MSI BIOS on the 2nd "performance" setting via the switch on it is 450w, with 108% power limit being the maximum in Windows, and 480w being the maximum in Linux and I presume Windows as well.

 

I did (notoriously) run the Galax 666w BIOS and folded on it 24/7 for like 4 months on that BIOS with a 2014-era Cooler Master V1000 (Seasonic) power supply. MSI includes a 12VHPWR dongle to three 8-pin PCI-E power. I was supposed to run three separate wires individually connected to the PSU to the connectors on that dongle, but I was stupid and lazy and daisy chained one of them. @bridgypoo pointed out the whole time that we should run another separate wire, but I was like "oh no my PSU is Seasonic and high quality and it'll be fine". These were NOT ATX 3.0 or whatever PCI-E 8 pins so their max wattage should be 150w, and 150w x3 is 450, just enough to meet the power limit of the card. I ended up frying cables and ports on the back of that PSU.

 

Anyway, I have a new Leadex 1300w unit since my Seasonic Vertex GX died, we got approved for RMA though. Its powering my folding just fine so far.

 

I do not use the 666w bios, but I do raise the power limit and overclock the card using an app in Linux. See picture:

 

 

Note the power draw in GreenWithEnvy. Folding takes 350w, to 400w max according to this. If I look at the Kill-a-Watt meter that just the box is plugged into, it's usually drawing 450-550 total. (And it's a 1300w psu) This new PSU is dual rail, the Seasonic Vertex we are RMAing was not. I don't fold on the CPU, just on the 4090.

 

Sorry that was so long.

 

 

I'm about to turn my rig off, take the glass off and measure. Thanks for this.

OK, cool, then I really wouldn't worry about it as long as you've made sure the plug is in all the way and there is no tension pulling on it, then should be good to go.

 

Hey thanks for that info @Slaughtahouse - was not aware they had that much room to work with. Still seems a lot for one cable with tiny pins, but I m not an electrical engineer. 

[neurotix]

We just checked the two 8-pin PCI-E connectors on the backside of the power supply unit. They are in all the way, just fine. With that and the within limits bend on the 12VHPWR on the GPU side, and it being on a separate rail, this thing should hopefully run my folding@home 24/7 365 OC'ed in Linux.

 

 

[J7SC_Orion]

9 hours ago, Sir Beregond said:

I guess if you were really nervous about it, are you really gaining much running 670W vs 450W on the 4090? Lack of power was never really the 4090's problem like it was with the 30-series. Hell 350W is basically within 5% of the performance of the card at 450W, so I have to imagine 670W is highly diminished returns. It's not 0, there are gains, but for an extra 220W seems minimal.

 

Now I'm not sure the difference in folding PPD vs gaming FPS, but something to think about if it really makes you worried.

 

All I know is the beefiest 12VHPWR cables are rated up to 600W, so you're probably maxing that out (and then some maybe) plus power draw from the PCIe slot. That's gotta be taxing the hell out of the PSU. If Nvidia sticks with this power connector, they really ought to put two of them on these higher wattage cards like the Kingpin or HOF cards did.

 

EDIT: Wrote this too early, didn't realize the 670W was @J7SC_Orion and peak power. What kind of normal power draw are you getting @neurotix? I thought you ran a 660W BIOS or something, but would be interested to hear.

 

 

 

 

On the 4090, some games - never mind certain benchmarks - can exceed a 'regular' 600 W...the real issue with 4090s though are the transient spikes (up to 700 W with both the stock 600 W and the Galax HoF 667 W vbios on mine). Transient spikes (ie. pic on the right) do not usually raise temps much as they are fleeting, but they certainly do not help with the 12VHPWR connector sensitivity issues. In MSFS 2020 for example, the 4090's base consumption is limited to 430 W, but transient spikes still exceed 650 W, and they tend to occur every few seconds.

 

All that said, on the left is a pic of my 3090 Strix OC w/KP 520 W PCIe connector. This via a Platinum-quality PSU with 16 gauge wires, but on one of the 6+2 PCIe, there was an issue with seating and the subsequent arc started to eat away at the card's connector area (purple arrow). Fortunately, I noticed in time and both the card and the PSU are still running fine these days in another system...

 

Conclusion : Caution and regular checking is a good thing, especially with the overworked 12VHPWR connectors in combination with ever increasing Watt consumption of top cards (5090 Blackwell for example is rumored to have twin / 'siamised' dies, each based on a modified 4090 die, so do the math on power consumption even with a reduced node and other improvements).

 

When I run the 4090 in gaming or benching, I do regular spot checks with a laser-pointed IR gun around the 12VHPWR connections, and also the near-by VRM on the back. Only takes a few seconds for some peace of mind 

 

 

 

Edited July 10 by J7SC_Orion

[pio]

On 10/07/2024 at 13:02, neurotix said:

I measured the length of the cable before the bend and it was about 1 3/4" so according to those charts, it should be ok (I hope). I am shocked@piois not in this thread making fun of me I fried another one bro, a Seasonic Vertex GX 1200w that was only a year and 3 months old. Lol.

I'm very much against the 12VHPWR connector, and well.......this is why.  Sorry you burned up another PSU.  Hopefully this one won't burn down, right?

@J7SC_Orion has a lot more experience with the 4090 and that 12vhpwr connector.  I'd be heeding his advice and regularly checking those connections.  If I'm not mistaken (I could be), the 12vhpwr plug DOES in fact have a limited number of times you can plug and unplug it before the connector is deemed broken.  I've heard there's similar "restrictions" on 6/8pin PCIe plugs, however I've never seen that restriction put into PSU warranty and user guides, nor have I seen that limitation tied to a cable like the PSU manufacturers have been doing with the 12vhpwr cables.  But every, EVERY modern PSU I've bought or received for review lately, has giant lettering stating there's something like 20 insertions for that plug and that's it.

I sincerely hope you get it all sorted out and stop having problems bro.  If that was MY CARD, I'd be putting stock vbios on it and run the thing at stock, to avoid future melting incidents.  I still want a 4090 eventually, but I get the feeling by the time those cards are down in the budget ebay categories, there probably won't be any left.  -_-

[J7SC_Orion]

11 hours ago, pio said:

I'm very much against the 12VHPWR connector, and well.......this is why.  Sorry you burned up another PSU.  Hopefully this one won't burn down, right?

@J7SC_Orion has a lot more experience with the 4090 and that 12vhpwr connector.  I'd be heeding his advice and regularly checking those connections.  If I'm not mistaken (I could be), the 12vhpwr plug DOES in fact have a limited number of times you can plug and unplug it before the connector is deemed broken.  I've heard there's similar "restrictions" on 6/8pin PCIe plugs, however I've never seen that restriction put into PSU warranty and user guides, nor have I seen that limitation tied to a cable like the PSU manufacturers have been doing with the 12vhpwr cables.  But every, EVERY modern PSU I've bought or received for review lately, has giant lettering stating there's something like 20 insertions for that plug and that's it.

I sincerely hope you get it all sorted out and stop having problems bro.  If that was MY CARD, I'd be putting stock vbios on it and run the thing at stock, to avoid future melting incidents.  I still want a 4090 eventually, but I get the feeling by the time those cards are down in the budget ebay categories, there probably won't be any left.  -_-

 

...you are right about warning re. too many plug-in, plug-out attempts with the PSU cable as the metal sleeves mating with the card connector's pins can loosen with repeated dismount / insertion. That leads to weaker contacts, more play, and potentially arcing. This is also why I recommended an IR gun with a laser pointer, you do not touch anything on the cable or connector with it, yet you get the info you want.

 

Below is a pic I took of my card's 12VHPWR connector two weeks ago when I mounted a new PSU with its native 12VHPWR cable. That connector has been handing 600 W+ since October 2022. You also see some minor remnants of the thermal putty on the other side and behind the connector on the GPU - this is after removing most of the old stuff and before adding a new layer. 

 

 

 

[neurotix]

Can you recommend a good IR gun, @J7SC_Orion?

[J7SC_Orion]

59 minutes ago, neurotix said:

Can you recommend a good IR gun, @J7SC_Orion?

 

..I got mine from 'Canadian Tire', but >this link is for Home Depot / US