TheaterFire

Buildzoid: Taking a look at Sapphire implementation of the 12VHPWR connector on the RX 9070 XT Nitro+

Posted by COMPUTER1313@reddit | hardware | View on Reddit | 112 comments

Reply to Post

112 Comments

swatop@reddit

Got a 9070 xt nitro+. Why? Because there was no other card available and I am meanwhile waiting for months for card to buy after my old 2070 broke down. Oh and I love the thrill of course. No seriously, in my country all RX cards sold out 3 mins after the release (despite being sold for +1200€ on the XT version and +1000€ for the non-xt one -> MSRP is a joke at this point) and I was lucky to even find this one. Ill hammer the connector in when the card arrives and hope for the best. Considering that it draws less power than a 5090 and 5080 it probably is not such a big deal after all. Have not read much about 5070s melting or 40 series cards which draw less power than the top models.
View on Reddit #50456819

Disastrous_Media_176@reddit

how is it performing? i hope that it is so far so good. lol
View on Reddit #50853699

Explosivpotato@reddit

Kinda makes me glad I ended up getting a card with 3 8 pin connectors. For the sub 400w draw the 9070xt has even when clocked to the gills, there’s so much overkill in that power delivery I’ll never have to worry about it. Theoretically I could push the better part of a kilowatt into this card before really having concerns.
View on Reddit #50475219

3G6A5W338E@reddit

I always default to Sapphire Nitro+, but they seriously fucked up this time. With that connector, I cannot recommend it. Any alternative that uses the known-good 8pin is automatically better.
View on Reddit #50357913

FlorrenEsseb-13579@reddit

Maybe the XFX Magnetic Air variant? That one has 3x 8-pin connectors.
View on Reddit #50368615

N1NJ4W4RR10R_@reddit

Looks like Powercolor's Red Devil has 3x 8 pins as well
View on Reddit #50405557

ThankGodImBipolar@reddit

Although Sapphire seems to get more attention online, as far as I’m aware, the Red Devil SKUs are normally just as good as the Nitro+ ones anyway. This is just an unforced error for no reason, and I’m kind of hoping that some of these cards melt so that Sapphire is discouraged from making the same mistake.
View on Reddit #50409656

Techhead7890@reddit

> Powercolour red devil ... normally just as good as the Nitro+ Eh, so these are different lines but the [HUB 9070 (non-XT) benchmark](https://youtu.be/gWIIA-a9Q9A) had the powercolour hellhound trailing noticeably behind the Sapphire Pure and XFX pro quicksilver. So I dunno, I'm a bit leery of Powercolor, I could be biased/wrong but I've always seen them as the inferior product.
View on Reddit #50454881

ConsistencyWelder@reddit

Maybe it's not so much the connector that is bad, but that they're pushing too much power through it. More than it can handle. Remember, the TDP of a 5090 is almost twice that of a 9070XT.
View on Reddit #50364564

Farren246@reddit

The entire point of the connector is to save space. But you can't save space and at the same time devote board space to power regulation. So the entire thing is flawed by concept alone. It cannot be made right.
View on Reddit #50368305

FlorrenEsseb-13579@reddit

Whoever thought compromising safety in the name of "saving space" was a good idea, I just wanna talk to them. [https://tenor.com/view/peter-griffin-meg-griffin-i-just-wanna-talk-to-him-put-that-gun-away-family-guy-gif-14677957](https://tenor.com/view/peter-griffin-meg-griffin-i-just-wanna-talk-to-him-put-that-gun-away-family-guy-gif-14677957)
View on Reddit #50368805

Techhead7890@reddit

Yeah it's crazy how they are just shrinking the PCB down more and more. I get that they wanted the airflow to cool it, I get that there are probably standard sizes and case widths so it's not like they can make it wider... But at the 5090 level, I swear they should just sell it with an AIO radiator. Modern problems require modern solutions etc.
View on Reddit #50454586

BatteryPoweredFriend@reddit

Yeah, it's pretty much like saying cars don't need things like seat belts or crumple zones, because you should just drive in a way to not have a car accident.
View on Reddit #50410108

3G6A5W338E@reddit

The primary complaint is that two or, if greedy, three 8pin connectors will do the job, and PSUs provide these, for a long time now. Thus there was no need to push such a problematic connector that few PSUs provide and most people will instead need adapters for.
View on Reddit #50364778

ragzilla@reddit

12v-2x6 adds sideband signals that don't take away from power delivery pins, also, it adds bidirectional sideband signaling so the *card* can tell the *PSU* if it's plugged in correctly or not. So future PSUs implementing the spec can disable power on the 12v-2x6 until the card says it's ok and ready to take power. Right now, the GPU has to self-police that.
View on Reddit #50394368

3G6A5W338E@reddit

Nice feature we cannot enjoy today. Instead, there are very real issues with that connector. In the future, the connector will get discarded due to its issues, before anyone can actually benefit from that feature.
View on Reddit #50442548

FlorrenEsseb-13579@reddit

Seriously, what wrong would've been committed using 3x 8-pin EPS-12V connectors.
View on Reddit #50368665

veritas-joon@reddit

because the vain people care about aesthetics.......and maybe a tiny sliver actually care about airflow through the case with the power cables disrupting case flow....maybe? I dont see it affecting case flow that much is you setup the fans in the correct placement
View on Reddit #50370516

FlorrenEsseb-13579@reddit

LMFAO they'll know what aesthetics are when their house has a big black dark spot and smells like burnt toast.
View on Reddit #50370897

Jeep-Eep@reddit

Yeah, at 330 watts... be better with 8 pin systems, but between the same safety margin as those and the wire receiving active cooling from the card, it's not a horror like Ada onward.
View on Reddit #50418346

reddanit@reddit

> Maybe it's not so much the connector that is bad, but that they're pushing too much power through it. If you want to be really anal about it, it's the spec that's bad. It is bad because it allows for components that follow it to burn up in fairly mundane scenarios. There are many different ways in which the spec could be improved and reducing its power rating is certainly one of them. Though out of all the possibilities, it's the option that is just about the furthest in "defeating the whole purpose of new connector in first place" department. IMHO the spec should either mandate load balancing, or as a cheaper option, monitor each pin and shut the whole thing down if any exceeds its rated current.
View on Reddit #50368128

ragzilla@reddit

>IMHO the spec should either mandate load balancing Why? When all the arguments against it involve contrived situations like "well if you cut 5 out of the 6 current carrying conductors...". Can you actually create a problematic current distribution if the system meets spec?
View on Reddit #50394483

reddanit@reddit

> Can you actually create a problematic current distribution if the system meets spec? The number of burned/melted cables and connectors is a pretty obvious proof that not only it's possible, but it actually happens on pretty regular basis. Detailed mechanisms of *how* this current imbalance happens are wholly separate thing. Which is interesting in its own right - some of them were even addressed by 12V-2x6 update to the spec. Ostensibly not all of them and knowing which ones are causing the failures in practice is pretty complicated - you'd need to do failure analysis on many of them and look for patterns.
View on Reddit #50397065

ragzilla@reddit

>The number of burned/melted cables and connectors is a pretty obvious proof that not only it's possible, but it actually happens on pretty regular basis. Are we talking about 12vhpwr, or 12v-2x6 here? Because they are 2 different PCB connectors, and as you know 12v-2x6 incorporates changes which prevents the partial insertion melts of 12vhpwr. So far for melted connectors in 12v-2x6 PCB headers there are **3** confirmed cases (and nearly as many confirmed frauds). 2 using worn cables from 4090 with potential problems (one the user admits wear was an issue, and the other was gold plated so there's potentially gold/tin galvanic/fretting corrosion issues). And one Astral 5080 which seems like a bathtub curve failure (no obvious cause).
View on Reddit #50398354

Bemused_Weeb@reddit

> Can you actually create a problematic current distribution if the system meets spec? Yes. Plug in the power such that it's at a weird angle that puts it under stress that becomes problematic over time. Alternatively, plug it in such that it looks right but there is actually a tiny gap. Cutting the conductors, though contrived, is a lot more reproducible.
View on Reddit #50395779

ragzilla@reddit

And you can do that on 12v-2x6, or 12vhpwr? Because you're aware, of course, that there is a difference between the two and that 12vhpwr had an engineering flaw which allowed what you described to happen, and the changes to 12v-2x6 were *specifically* to address those flaws. I was more thinking of someone providing a spec-compliant terminal resistance scheme which results in problematic load balancing.
View on Reddit #50396088

FlorrenEsseb-13579@reddit

I'd just argue the cheaper option should not be an option at all, it should just straight up be illegal, as then you'd get a card you can't even play games with without your card shutting down every time it hits the power limit. It's just bad design overall.
View on Reddit #50368734

RandomGuy622170@reddit

You and I are very similar. I was hoping to grab a Nitro or white Pulse but can't justify it with that shitty connector. It's a shame they went that route.
View on Reddit #50432331

COMPUTER1313@reddit (OP)

TLDR: Sapphire uses the same flawed 12VHPWR connector design on their RX 9070 XT Nitro+.
View on Reddit #50350585

JuanElMinero@reddit

I just checked and they use it on both the 9070 and 9070 XT Nitro versions, obviously being less of a potential issue for the 9070. Alas, the PULSE and PURE lines still exist, both featuring 2x8pin connectors
View on Reddit #50351236

hhkk47@reddit

Pulse it is then. I've gone with Sapphire Nitro+ cards for my last 3 GPU purchases because of how well they are (or at least seem to be) built, but the completely unnecessary switch to 12VHPWR is a deal breaker.
View on Reddit #50453188

Lukeforce123@reddit

The pure has the white electronics tax™
View on Reddit #50367333

kikimaru024@reddit

Not true, it also has a 1.6% overclock!
View on Reddit #50367706

Average_C_Programmer@reddit

Overclock tax
View on Reddit #50422406

RedofPaw@reddit

Why would nVidia do this?
View on Reddit #50452212

battler624@reddit

It cant go past 20A on 1 pin due to the 2 fuses if i'm understanding correctly.
View on Reddit #50351025

RealThanny@reddit

You are not understanding correctly. Those fuses protect two 12V rails as they leave the single 12V input rail.
View on Reddit #50448512

mtthefirst@reddit

The fuses are after the connector. It's for the power stage protection. If both are 20A each, so the max current that can pass through the connector assume that both 12V power circuit share the same load is 40A (480W). The failure can still happen to the connector, if one pin carry all those current.
View on Reddit #50351852

ragzilla@reddit

40A is past card TDP (480W vs 304W), so unless the VRM is misprogrammed it'll never pull more than 25.3A, and the contrived situation to get to that point is pretty unlikely unless you're cutting circuits for *aesthetics*. Roman's cable was misbalancing to 22A on a single circuit and he didn't even know he had a problem. 25.3A on an in-spec terminal is only 3.9W of dissipated power, which (assuming ground balances evenly as matches my observations) is going to be within the thermal envelope of the connector, as ground will only be contributing 0.64W of power across 6 pins. If you cut some more of the ground pins you could create yourself a problem though, or the hotspot on one terminal could be a problem especially if it got to thermal runaway (which isn't as likely with the connector in airflow).
View on Reddit #50393902

Jeep-Eep@reddit

Yeah, this wiring would be frightful on a 9090XT or UDNA one, but for 330 watts? Acceptable.
View on Reddit #50418478

ragzilla@reddit

I think it's fine up to 600W TDP but still need to do a larger writeup in one place. People are laying a lot of 12vhpwr issues at the feet of 12v-2x6.
View on Reddit #50418797

Jeep-Eep@reddit

I personally would not take the spec beyond 375 watts. A 9070XT is fine, but it's the biggest card I'd feel comfort with one of those plugs as they're implemented now.
View on Reddit #50419022

ragzilla@reddit

I've been meaning to write some code to sweep the potential values in the connector to look at design thermals (I'm 99% confident the design thermal load is 5W but want to see what varied spec-violating combinations could do, while meeting <=6mOhm LLCR requirement). And then maybe I'll buy some PTC elements and go push 600W through 3 circuits for an extended period (likely pushes the thermal envelope a little if the 3 circuits are at 6mOhm terminal resistance). Heck maybe I'll bin some terminals and go run 600W through 2 terminals, because 2 at 25A seems doable if they're <2mOhm.
View on Reddit #50422997

Jordan_Jackson@reddit

The card could still pull more than that through just one of the pins, if the right circumstances exist. If it does, the fuse is supposed to blow. Doesn’t help with potential melting connector issues though. Using this connector and not improving on it, is a dumb decision by Sapphire.
View on Reddit #50368935

karlzhao314@reddit

>The card could still pull more than that through just one of the pins, if the right circumstances exist. If it does, the fuse is supposed to blow. That is not correct. The fuse doesn't protect the connector *whatsoever*. The way the power delivery is designed has all six conductors of the connector merging to one power plane, passing through a pair of shunts, and *then* being split off into two separate power rails, each with their own fuse. The VRM phases are distributed between the two power rails (so, 5+1/5+2 for each rail, if Techpowerup's board analysis is to believed). The purpose of the fuses is to protect the card in case one of the VRM phases blows. That will also blow the fuse almost immediately, which ideally means that the rest of the card will be protected and only that VRM phase will need to be replaced.
View on Reddit #50378773

Jordan_Jackson@reddit

Reread my post. I never claimed that it would protect the actual connector. Maybe I worded it a bit funky.
View on Reddit #50384414

karlzhao314@reddit

You said, >The card could still pull more than that through just one of the pins, if the right circumstances exist. If it does, the fuse is supposed to blow. It does not do that.
View on Reddit #50384482

Jordan_Jackson@reddit

And that does not say that it’s protecting the connector. And yes, the card pulls the power it needs through both the connector and the PCIe slot. Reread my post and learn reading comprehension
View on Reddit #50384670

karlzhao314@reddit

If the fuse blows when the card draws 20A through one pin as you claimed, then by definition, the purpose of that fuse is to guard against overcurrent on the connector. It doesn't do that. Either way, your claim is wrong, regardless of whether you're arguing semantics about whether you stated it "protects the connector" or not. The purpose of the fuse is not to guard against overcurrents on the connector pins, but rather to guard against blown power stages. Anyway, I'm done here. Have a good day, and thanks for the downvotes.
View on Reddit #50385028

Jordan_Jackson@reddit

Go away. I never claimed that it protects against the connector. Show me where I said those words. If anyone is arguing over semantics here, it is you because you seem to be the type that likes to correct people. Go back to where you came from dude.
View on Reddit #50385261

Quatro_Leches@reddit

its even hard to fuse this, usually you'd use a fuse with overcurrent headroom of like, 25% at least?. the fail headroom for VHPWR is less than that lmao its a flawed connector design, simple ohms law. you need bigger gauge contacts and bigger gauge wire. or more wires.
View on Reddit #50352589

karlzhao314@reddit

I don't like the idea of fusing every individual pin either way, unless you have some way to shut the card down as soon as any one of the fuses blows. Because otherwise, when one fuse blows, the current gets distributed to the remaining five. Also, if there are going to be non-resettable fuses used on a card, I'd much rather have them only blow when something truly *catastrophic* happens to the card. A blown power stage is catastrophic, and warrants the card rendering itself unusable until a tech replaces the power stage and the fuse. I wouldn't necessarily call an unbalanced current automatically catastrophic, especially when it can be remedied by the user unplugging and plugging in the power cable a bit better. The real solution here was load balancing, which I have no idea why everyone switching to 12VHPWR ditched.
View on Reddit #50385876

ragzilla@reddit

1. for in-spec systems, you can demonstrate active load balancing is unnecessary 2. active load balancing increases card complexity and can reduce reliability at higher TDP (since you now have more actively controlled things which can go wrong and under-volt the card during voltage transients) 3. doing "perfect" load balancing on 12v-2x6 requires a part which nobody makes yet, a 6-channel high side current shunt monitor.
View on Reddit #50392831

Acrobatic_Age6937@reddit

i still think the balancing should happen on the PSU side
View on Reddit #50401079

ragzilla@reddit

You can't. Current limited supplies do so by lowering voltage, and if you did that in a system like this it just swings the load to the new highest point of potential. You can only steer and balance on the consuming side.
View on Reddit #50401940

Acrobatic_Age6937@reddit

i agree that it might cause problematic interactions with SMPS circuits. A simple electronic fuse might be all it takes. By threatening the gpu/ the consumer with instant shutdowns if they don't balance properly, they have to do it. If I've learned one thing in the industry than that people prefer the 'easiest solution', the only way to prevent shortcuts is by making the right solution the easiest/only option.
View on Reddit #50402709

ragzilla@reddit

In an ideal world, we'd have a 6 channel current shunt monitor and NV could do a 6 supply rail VRM topology that would never overcurrent a single pin. Unfortunately, that doesn't exist, and I'm also guessing there's other reasons they went to single supply rail topology (such as, higher TDP mean your VRM is working even harder to not undervolt the chip, and split supply rails make that problem substantially more difficult because instead of just driving all VRMs in a rail a little harder, you have to make a decision as to which rail to drive harder and that takes time, time which might result in enough Vdroop to drop the chip).
View on Reddit #50404233

Acrobatic_Age6937@reddit

Still when the risk is a potential fire, i think it's not too much too ask. No one would sign off a house without over current breakers. imagine explaining the electrician 'well, I believe it's the consumers responsibility to not overload the wiring' :D I with there was an actual lawsuit about this. I'm somewhat sure the psu manufacturer wouldnt get out of it scot-free
View on Reddit #50404817

ragzilla@reddit

>Still when the risk is a potential fire, i think it's not too much too ask. No one would sign off a house without over current breakers. imagine explaining the electrician 'well, I believe it's the consumers responsibility to not overload the wiring' :D For ground and short fault protection, there's protection in the PSU. Conductor thermal protection is at the load. You'll see similar protection schemes in houses any time there's an inductive load (i.e. a motor). Chances are the breaker for your air conditioner is oversized for the conductors because we trust the motor at the AC to protect the wires- just like a GPU. But fire is unlikely based on the materials specced in the connector housings.
View on Reddit #50407307

Acrobatic_Age6937@reddit

the standard here is that you have to have a breaker that limits the current to what the weakest link can handle, the device you plug in being a variable are obviously excluded from this. Which in reality means if you have 1.5mm2 cabling you need to protect it with a 16A fuse. Outlets ofc. have to be capable of handling that as well. > Chances are the breaker for your air conditioner is oversized for the conductors because we trust the motor at the AC to protect the wires- just like a GPU. Wouldn't be to code here. How could you trust a random device to not develop a fault?
View on Reddit #50411086

ragzilla@reddit

>the standard here is that you have to have a breaker that limits the current to what the weakest link can handle, It's likely you don't, unless you live in a jurisdiction with no inductive loads, or where they **really** like to waste copper. NEC articles 430 and 440 cover some of this. >How could you trust a random device to not develop a fault? Same way you trust the breaker not to develop one.
View on Reddit #50414327

Acrobatic_Age6937@reddit

> Same way you trust the breaker not to develop one. the breaker is far less likely to fail. And at the end of the day two things have to fail in this scenario. > It's likely you don't, unless you live in a jurisdiction with no inductive loads, or where they really like to waste copper. NEC articles 430 and 440 cover this in US jurisdictions. I'm from Germany. In my apartment every circuit has 230V and a 16A breaker. If you need 32A on one phase you would have to install a 6mm2 cable e.g. for induction cooktops. At that point 400V becomes somewhat more common though.
View on Reddit #50417268

ragzilla@reddit

>the breaker is far less likely to fail. And at the end of the day two things have to fail in this scenario. The thermal trip in your breaker is actually the very first thing to go over time, because it's mechanical in nature (bimetal strip). >I'm from Germany IEC 60364-4-43 is the applicable model code here, specifically down in part 432 which is where it starts dividing into the multiple types of overcurrent. Inductive loads like motors have high inrush currents which mean you need to oversize the breaker, which by conventional wisdom says you oversize the conductor as well. 435.2 is particularly applicable talking about this: >435.2 Protection afforded by separate devices The requirements of Clauses 433 and 434 apply, respectively, to the overload protective device and the short-circuit protective device. The characteristics of the devices shall be coordinated so that the energy let through by the short-circuit protective device does not exceed that which can be withstood without damage by the overload protective device. This clause is what would let you oversize a breaker in the panel, based on a load's own internal protection. The specifics are buried up in 433.3.1(b) (because the wiring section between the breaker and the thermal protection at the motor makes this applicable) which lets you ignore the overcurrent requirement at the distribution location, because the motor's thermal protection then provides 433(.2.1) overload protection. So, all the panel breaker has to provide is short/ground protection under 434, which makes it a 423.3 installation at the panel: >432.3 Devices ensuring protection against short-circuit current only A device providing protection against short-circuit current only shall be installed *where overload protection is achieved by other means* or where Clause 433 permits overload protection to be dispensed with. Such a device shall be capable of breaking, and for a circuit breaker making, the short-circuit current up to and including the prospective short-circuit current. Such a device shall satisfy the requirements of Clause 434. IEC code's a little weird sometimes, personally, I prefer NEC as it keeps more of it in one document. Inductive load may have thrown you for a loop there, the more frequent inductive loads would be motors (commonly found in HVACR equipment). An induction cooktop would likely use a similar protective scheme though.
View on Reddit #50422571

Acrobatic_Age6937@reddit

> This clause is what would let you oversize a breaker in the panel, based on a load's own internal protection. I think there's a misunderstanding here. The circuit breaker will rarely factor in the load (at least not in most private installations). It's designed to protect the cabling. By 'oversizing a breaker' you mean the breaker has a higher tripping point than what the load can safely consume, right? That would assume you know the load when installing the breaker. > So, all the panel breaker has to provide is short/ground protection under 434, yes i agree with that. The issue here isn't that the gpu itself catches fire, but the known wiring/ connector gets stressed beyond it's specification. That's in of itself not a short circuit, but in a house wiring would still trigger the breaker. I guess it would be more sensible to call it an 'wiring overload circuit breaker'
View on Reddit #50425573

ragzilla@reddit

>It's designed to protect the cabling. *Only* if it's being used for *overcurrent* protection. As you see from the IEC references, this is not always the case, overcurrent *can* be provided elsewhere in the circuit, so long as the portion between the overcurrent device and the current source has short circuit protection.
View on Reddit #50427024

ragzilla@reddit

>its a flawed connector design, simple ohms law. you need bigger gauge contacts and bigger gauge wire. or more wires. It takes a pretty contrived terminal resistance scheme to create a problematic current distribution. The kind of thing you only see if you have substantial mechanical wear on the connector, or if you have a system that's exposed to galvanic/fretting corrosion.
View on Reddit #50392625

binary_agenda@reddit

Is this connector part of the ATX 3.0 standard? Or just some bullsh*t Nvidia made up?
View on Reddit #50373638

ragzilla@reddit

12v-2x6 is a PCI-SIG standardized connector, which is integrated by reference into ATX 3.1. ATX 3.0 integrated the previous 12VHPWR connector, which appears the same on the outside, but there are differences in the pin lengths inside the PCB connectors. H+/H++ stamps on the PCB header are used to externally identify them as 12VHPWR/12v-2x6 respectively.
View on Reddit #50394128

binary_agenda@reddit

If it's in the standard I don't see how AMD could make a substantially improved version. Does the improved pin length resolve any of the melting issues?
View on Reddit #50394756

ragzilla@reddit

The pin length changes address the partial insertion issue which plagued 12VHPWR (see GN's video on this for an example). It requires the connector be fully inserted for the sense pins to make contact *after* the power pins contact, and until the sense0/sense1 pins are grounded by the PSU the card isn't permitted to draw power. Before on 12vhpwr the sense pins were longer than power, so they'd make contact first so you could have power hanging by a thread but good sense contact.
View on Reddit #50395351

binary_agenda@reddit

Good stuff, thanks!
View on Reddit #50399923

plantsandramen@reddit

Way to go sapphire. I was going to buy the nitro+, I made extra room in my case. Now I'm going PowerColor.
View on Reddit #50366966

Farren246@reddit

And no one was surprised.
View on Reddit #50366913

ET3D@reddit

Thanks a lot for the tl;dw.
View on Reddit #50366735

salcedoge@reddit

That’s a dealbreaker imo considering the AMD cards are having issues with power as well, you really want to minimize the risk at all cost
View on Reddit #50352131

NA__Scrubbed@reddit

I mean, yeah I'd rather they went with 3 8-pin, but how many 4080 or 5080 cards have burned? 1? 10? The fuses at least mean any failures will almost certainly be easy to repair. You have excellent anti-sag and anti-warping features in this card, user swappable fans, steel frame, pre-installed PTM 7950, etc etc. In all likelihood this card will last much longer than other models before it needs any kind of repair.
View on Reddit #50384129

Jeep-Eep@reddit

Yeah, it ain't as good as it ought to be, but it's effectively derated within the safety margin of the spec, and Sapphire warranties allow undervolts anyway. Now, a Big UDNA one or a 9090XT would be another story.
View on Reddit #50397326

ragzilla@reddit

1 5080 card (unknown cause), 2 5090 cards (one the user admits was likely cable wear, and the other the user had a gold-plated moddiy cable from their previous 4090 so there's potentially years of galvanic/fretting corrosion on the terminals). Techtubers not really understanding the difference between 12vhpwr and 12v-2x6 doesn't help either.
View on Reddit #50395079

COMPUTER1313@reddit (OP)

And all of that happened within 3 weeks of the launch and with very limited supply.
View on Reddit #50395963

Slyons89@reddit

Knowing this has convinced me to go with an xfx mercury or Asus TUF, it’s too bad because the sapphire card is gorgeous. I was hoping they could do better with the new connector. Makes me wonder if 6 pins to one power plane is actually part of the official spec now, because every card uses that , even the ones that have PCB space and high enough profit margin on the card to allow for currency balancing components.
View on Reddit #50352314

Sh4rX0r@reddit

It cannot be part of the official spec because the 3090 Ti was not like that.
View on Reddit #50355643

mapletune@reddit

specs evolve all the time. that's how we got the new "12V-2x6 (H++)" in 2023 which is an update from the original introduced in 2022. the connector used in 3090 Ti isn't part of the standard, it was nvidia proprietary.
View on Reddit #50356494

COMPUTER1313@reddit (OP)

Oh, and while it is physically compatible, you aren't supposed to use the old 12VHPWR cable with the new 12V-2x6 ports or risk burning the cable. Compare that to USB where if you plug a USB 2.0 cable into a USB4 port, there's no risk of the cable catching on fire.
View on Reddit #50375842

ragzilla@reddit

There is **no** difference between a cable for 12vhpwr and a cable for 12v-2x6. The differences are inside the PCB headers.
View on Reddit #50394605

COMPUTER1313@reddit (OP)

I’m confused now. About a month ago, someone posted a picture of them using an old cable from their RTX 4xxx card with the 5090 and it melted. Everyone was blaming the user for using the wrong cable.
View on Reddit #50395750

ragzilla@reddit

Those people are dumb (there's a lot of it on reddit). There are confirmed 2 melted 5090 connectors so far: 1. The brazilian/portugese guy, who admits his cable was heavily worn and likely contributed to the failure. Cable was previously used on 4090. 2. The german guy who sent his card and cable to der8auer. This was a gold-plated moddiy cable which was previously used on their old 4090. My hypothesis for this one is that during the time it was used on the 4090 the terminals were degrading due to galvanic and fretting corrosion due to mixed mating metals and the reseat to 5090 and subsequent TDP increase pushed it from "almost a problem" to "problem". [About the 12V-2x6 connector that replaces the 12VHPWR](http://jongerow.com/12V-2x6/) is jonnyGURU's commentary on the 12v-2x6 >Also, it needs to be pointed out that there is no change to the actual cable's connectors and terminals. For the new connectors to work on the GPU side, and in many cases the PSU side, the cable has to remain the same. Despite this, some marketeers are calling their cables "12V-2x6" despite being the same as a 12VHPWR cable. While some may say this is unscrupulous behavior to make consumers think that somehow a 12V-2x6 cable is more advanced than a 12VHPWR cable, I have to say that if the entire industry just stops using the 12VHPWR name on the cables and just uses 12V-2x6, there will be less confusion in the market.
View on Reddit #50396486

Slyons89@reddit

According to major PSU suppliers (Corsair, seasonic), that is not true, the cable side is identical and compatible with both 12Vhpwr and 12v-2x6 connectors on GPU.
View on Reddit #50377869

karlzhao314@reddit

I've seen the PCI-SIG documentation regarding the connectors, and that's true, the cable side plug specs have not changed. Only the board/optionally PSU side connectors have changed with the 12V-2X6 spec. If PSU manufacturers had never introduced "12V-2X6 cables", there would be a ton of confusion from consumers about whether their old 12VHPWR cables were compatible with this new 12V-2X6 standard. Now that they *have* introduced "12V-2X6 cables", there's a ton of misinformation from consumers about how you *must* use a 12V-2X6 cable with your 12V-2X6 GPU. I honestly don't know which is worse.
View on Reddit #50382205

Sh4rX0r@reddit

Hmm, got it, thanks. They would have been really dumb to make it worse and spec it out like that, but you never know, lol.
View on Reddit #50356621

Slyons89@reddit

3090 Ti came out before 12Vhpwr spec was finalized.
View on Reddit #50369752

Yebi@reddit

> Makes me wonder if 6 pins to one power plane is actually part of the official spec now It's almost like you watched the video
View on Reddit #50363880

MonoShadow@reddit

I head an opinion 12VHPWR connector on RTX 3000 series has load balancing only because they were made with 8 pins in mind and switched to 12 pin after the fact. Back then nVidia allowed for 8pins and almost everyone went with it. Maybe they considered it themselves or some other reason. If this is true, then it had 3 resistors because there was an option to put 3 8 pins on the board at any point in time. When they went all in on 12 pin there was no need for this load balancing so they got rid of it. So "works as intended" and the only reason whey there were no issues with 12pin is 8pin design carryover.
View on Reddit #50368445

karlzhao314@reddit

That doesn't really align with the fact that the 3090 FE and 3090 Ti FE also had load balancing, and they were designed explicitly for the 12-pin. In fact, what makes even less sense is that the 3090 Ti FE PCB was essentially a "trial run" for the 4090 FE PCB. If you compare them side by side, they're 95% identical. The only differences are the lack of SLI fingers, a *couple* of different ICs, and...the fact that the power delivery no longer has independent power rails or load balancing on the 4090. I still feel like there *must* have been some engineering reason to remove load balancing from the 4090 design that goes beyond simply "cost savings", because otherwise it makes no sense that they explicitly went and *redesigned the board* between the 3090ti and 4090 just to remove load balancing. I just have no clue what that reason could possibly be, and it's completely fucked everything else up in the meantime.
View on Reddit #50379334

ragzilla@reddit

>I still feel like there *must* have been some engineering reason to remove load balancing from the 4090 design that goes beyond simply "cost savings" Load balancing is a significant complication, and as you push the system closer to the TDP limits, a multi-rail load balanced VRM is more likely to undershoot on a phase and crash your card.
View on Reddit #50394917

vegetable__lasagne@reddit

8 minute 50, is that the shortest video he's ever made?
View on Reddit #50360733

Farren246@reddit

Could have come in at 3 seconds: "Of course it's flawed."
View on Reddit #50368389

Ramongsh@reddit

Argumenting proberly takes time.
View on Reddit #50378561

Farren246@reddit

I'm not arguing, I'm agreeing. It's damn short.
View on Reddit #50387526

Ramongsh@reddit

No, I meant that Buildzoid has to argue for his point-of-view, and that takes longer than just making a statement.
View on Reddit #50387744

Farren246@reddit

Ah I see. Though Nvidia successfully argued for that point of view in designing the cable with more current passed over smaller wires, and a mentality of "make everything smaller" that precludes any company from implementing proper voltage regulation on the PCB (if they're going to implement proper checks then the PCB will take up so much room they could have just stuck to 8-pins).
View on Reddit #50388969

EdzyFPS@reddit

The 9070xt can pull 400w+ that's not good. Definitely avoid the Nitro+.
View on Reddit #50360545

SirActionhaHAA@reddit

It doesn't. The card pulls 300+w max in all games. If you're thinking of hwu numbers those include the cpu. https://www.techpowerup.com/review/sapphire-radeon-rx-9070-xt-nitro/41.html https://youtu.be/yP0axVHdP-U?t=1385
View on Reddit #50363194

kaibaceto@reddit

But power spikes are as high as 450w
View on Reddit #50367262

Muppet1616@reddit

Which isn't a huge deal for overheating of the power cables (it is for the power supply requirements). A transient spike of a few milliseconds to 500w is so short it doesn't put out enough heat energy to heat up anything.
View on Reddit #50367680

EdzyFPS@reddit

In the hardware unboxed review you can see 400w+ being pulled. I guess their numbers must be wrong.
View on Reddit #50363256

Az1234er@reddit

> I guess their numbers must be wrong. They use power draw for the whole system, which makes sense, since it takes into account that some load can be higher on the CPU. So the combined value is more useful
View on Reddit #50364864

SirActionhaHAA@reddit

Yes and those numbers are for gpu+cpu. On their charts a 3090 pulls almost 600w, and the 7900xtx almost 500w
View on Reddit #50363654

uzzi38@reddit

To be clear, the max with the biggest factory OC model and power slider in drivers maxed out is 375W. No RDNA4 cards should be pulling more than that sustained over a long period of time (of course, transient spikes can still happen).
View on Reddit #50364472

vegetable__lasagne@reddit

If no one is load balancing power on a 5090 nobody would do it on a card that costs a third as much.
View on Reddit #50362167

oloshh@reddit

Been purchasing pulse cards always and will continue to do so
View on Reddit #50364305

wusurspaghettipolicy@reddit

Does the connector on this card with the sensing pins disabled like the 3090Ti mean this is generally safer? Sensing pins are a reason to shut off the card causing 100% fans, black screen if I am understanding this. Are they using the actual sense pins for this connection?
View on Reddit #50354514

Techhead7890@reddit

The "12V High Failure Connector" quip was such a dunk. Love it.
View on Reddit #50351204