And other people are using made-up statistics to justify it is bad. I agree it is a sucky connector but stick to facts, not made-up statistics.
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
RTX 5090 FE Molten 12VHPWR
- Thread starter erek
- Start date
And other people are using made-up statistics to justify it is bad. I agree it is a sucky connector but stick to facts, not made-up statistics.
- Joined
- Dec 19, 2005
- Messages
- 18,477
- Joined
- Dec 24, 2008
- Messages
- 2,269
I mentioned earlier in this thread to one of your posts, but you might have missed it. Do you have any concern that equalizing currents across all wires with varying resistance actually increases heat generated in the connector? It's nice if you separate the vrm inputs that you can more easily monitor the voltage at each pin, so you could, in theory, shutdown before any damage is done, but electrically/thermally it doesn't seem to help the connector situation.
edit: I suppose it's nice that you can keep the max current below the connector spec, but I doubt that max current is actually the best measurement for a connector anyways, some combination of max power/ max temp of the connector makes more sense.
Yes, the purpose of splitting the input currents is to divide the current across the connector spec and keep each pin to a safe, non-overloaded current at max load. By definition, doing this minimizes the heat generated in the connector. BTW, what fundamentally limits the current through any connector/conductor is the heat generated by its resistance in response to the current -- the temp has to stay low enough for the connection to not degrade during use.
I believe the connector fail mode here is a slight imbalance in current causes a pin to get too hot because it exceeds the connector's spec. It's resistance degrades (increases), which makes it get hotter. This increase in resistance also moves to other pins as they take more load, and then they get hotter. Eventually it melts the plastic. The added heat is also not good for the pin and socket plating. The failure shows itself in the connector, but the defect itself is the PCB overloading the connector, which is why the littlegreenspaceman's posts of melted 8 pin connectors is absolutely meaningless without context of the connector quality and details about the video card they were attached to.
Now there is a max connector current per pin spec, and an absolute maximum rating for the connector current per pin spec. Good engineering practice means you run these pin currents well below absolute max when the video card is drawing max load and the system is at max temperature. If one does this, the connector will essentially last forever and there is a great tolerance for lower spec connectors that may find there way in there.
Shunts (small resistors) are used on these feeds to monitor the current. The BIOS and board firmware maintain these currents to a safe level. In Afterburner, the board power is shown as a percentage. Default is 100%, the TDP rating of the card and there is usually an allowed overage.
The Titan XP is a 250W card, so 100% allows the card to draw 250W. It can be increased to 120%, which is 300W. It has one 8 pin and one 6 pin connector rated at 150W + 75W = 225W, the rest of the power comes from the PCIe slot itself (75W).
The EVGA 1080TI FTW is a 280W card, so 100% is 280W. These cards have two different maximums depending on the BIOS selected, either 118% (330W) or 128% (358W). The higher power BIOS runs the fans faster. This board has two 8 pin PCIe connectors, so 150+150=300W. @358W, the board is going to pull 58W from the PCIe slot.
Buildzoid's PCB analysis videos on youtube are very good at describing these power supplies, the ones for the EVGA 1080TI FTW and 1080TI FE are particularly good and will correlate with what I've said above. He also has one for the original Titan Pascal, and you will see why that one is poor. It appears the Titan XP uses the 1080TI FE PCB.
One other things he goes into are the conductive resistances of the MOSFETs. Just like for conductors, those resistances are what limit the power supplies capability because temps have to stay low enough not to fry the components. They guy is very good at explaining at a fairly laymen's level how these power supplies work. In reality, switching power supplies are incredibly complex to design -- it is specialization in electrical engineering itself.
The power supply designs on the 1080 TI FTW3, 1080TI FE, and Titan XPs are masterpieces. I look at the 4090 and 5090 and react with "What the hell happened?"
- Joined
- Dec 24, 2008
- Messages
- 2,269
I think you misread what I wrote. Splitting the current equally across connectors will INCREASE the heat produced in the connector.
Example below. This example has 4 12V pins. The GPU is assumed to be consuming 400 watts. On the left is a "balanced" current example. On the right is no balancing, just 12V pins all connected to a single net on the GPU. Green columns is resistance in each pin. For this scenario I set one pin to have 0.033 ohms resistance, while the rest have 0.01 ohms. See the orange highlighted per pin waste heat, and all pin waste heat. I have more scenarios, in every single one, balancing the current causes the connector to heat up MORE.
Attachments
Last edited:
TLDR: If you want to measure the current going into one of these multi-conductor connectors/wires you have to use a DC clamp on ammeter, ideally capturing all the wires.
My previous post was already quite long, but I've been pondering posting this experience here for a while because it began my researching the issues here a lot deeper than I had experienced. The lesson is directly applicable to what is going on here.
I've got several 5800X3D Titan XP SLI rigs, one does train simming and FPS games, one is for flight sim, and the other is for driving sims. Needed a power supplies to drive them and I'm cheap. Long before the current pricing madness I sourced several used 800W Delta power supplies that came out of high end HP workstations.
These power supplies had a non-standard 6 pin connector on them to power a second CPU board in the workstation. 3 12V feeds and 3 grounds. It also has an 8 pin EPS12V connector with two feeds on it (18A + 18A). I wanted to adapt the EPS12V connector to a pair of PCIe 8 pin connectors for the video cards. I wanted to convert that non-standard 6 pin connector to drive the motherboard's EPS12V socket with six pins instead of 8.
Ok, so how much current was the EPS connector on the motherboard pulling at max load? I know, I'll splice in an ammeter on one of the 3 power conductors and multiply that by 3. Only, when I did the test, the ammeter read nearly zero. Wut? It can't possibly be zero. (That still I posted from debau8er's video showing the two glowing wires in the 12VHPWR harness is that video card doing the exact same thing.)
The problem is the ammeter has a sizable shunt in it making the ammeter path much more resistive and the current bypassed the meter and took the easiest path on the other two wires. Fortunately, decent clamp on DC ammeters were available on Amazon, like $30-$40 and problem solved. For my application, 3 wire pairs for the EPS12V is fine and the clamp on meter was well above zero at about 6A or so max load, so 2A per wire/contact.
This experience really sent home the lesson about these types of power connections being *very* sensitive to individual resistances. That shunt in the ammeter above is milliohms in size. Big compared to wire, but not really big. Stuff like this can usually be ignored, but not in high current power systems.
My previous post was already quite long, but I've been pondering posting this experience here for a while because it began my researching the issues here a lot deeper than I had experienced. The lesson is directly applicable to what is going on here.
I've got several 5800X3D Titan XP SLI rigs, one does train simming and FPS games, one is for flight sim, and the other is for driving sims. Needed a power supplies to drive them and I'm cheap. Long before the current pricing madness I sourced several used 800W Delta power supplies that came out of high end HP workstations.
These power supplies had a non-standard 6 pin connector on them to power a second CPU board in the workstation. 3 12V feeds and 3 grounds. It also has an 8 pin EPS12V connector with two feeds on it (18A + 18A). I wanted to adapt the EPS12V connector to a pair of PCIe 8 pin connectors for the video cards. I wanted to convert that non-standard 6 pin connector to drive the motherboard's EPS12V socket with six pins instead of 8.
Ok, so how much current was the EPS connector on the motherboard pulling at max load? I know, I'll splice in an ammeter on one of the 3 power conductors and multiply that by 3. Only, when I did the test, the ammeter read nearly zero. Wut? It can't possibly be zero. (That still I posted from debau8er's video showing the two glowing wires in the 12VHPWR harness is that video card doing the exact same thing.)
The problem is the ammeter has a sizable shunt in it making the ammeter path much more resistive and the current bypassed the meter and took the easiest path on the other two wires. Fortunately, decent clamp on DC ammeters were available on Amazon, like $30-$40 and problem solved. For my application, 3 wire pairs for the EPS12V is fine and the clamp on meter was well above zero at about 6A or so max load, so 2A per wire/contact.
This experience really sent home the lesson about these types of power connections being *very* sensitive to individual resistances. That shunt in the ammeter above is milliohms in size. Big compared to wire, but not really big. Stuff like this can usually be ignored, but not in high current power systems.
- Joined
- Dec 24, 2008
- Messages
- 2,269
Err... Im assuming this is unrelated to my post directly above yours?
Okay, I see where you are coming from, but I think the answer is we don't care nearly as much about waste heat as we do about not overloading the individual connector pins. In your example, the waste heat difference is only ~1W, in a 400W system that's only 0.25%. As long as the pin is rated for more than 10.178A, all would be well. BTW, 400W across 4 pins sounds like a lot. I suspect with more pins, losses from balancing would be even less.
Part of the calculus of the design would be looking at exactly what you are looking at here and incorporating it. In fact, it would guide, among other things, how crappy the connector can be. Ideally, you want a crappy connector as crappy means cheap. The more you can tolerate cheap, the lower your cost.
You could use an expensive and finicky power connector and one power phase. For literally pennies more, you could implement several input power phases on the PCB and use a cheap ass connector. I would chose the latter and Nvidia used to.
No it wasn't, you posted yours while I was writing it
. Somewhat off topic excursion on the same general topic. Perhaps I should have replied to my own post.- Joined
- Dec 24, 2008
- Messages
- 2,269
So in that example power disappated in the connector went up from 3.4 watts to 4.4 watts. I would say thats not a trivial percentage increase, especially since the connector is the problem. I would not feel comfortable current splitting onto pins that have unequal resistance, even if the current is below the max curent spec as per the datasheet
Quibble, the card probably* should only be pulling 65W from the slot. Unless it's also taking the full 10W for 3.3V from the slot, it's slightly overloading something somewhere.
* Years ago, when some AMD cards were found to be drawing >100W from the slot and burning up cheap mobos, someone tested slot power draw. IIRC AMD was taking the 10W of 3.3, while NVidia had decided it wasn't worth it. That was probably a decade ago though, so I wouldn't take it as gospel for current cards.
The 3.3V supply was only in PCIe though to aid parallel PCI cards porting to the 1.0 standard by giving them the power the legacy board was designed for and only requiring a bridge chip be added between the main card and the slot. Other than trying to get a bit more power without needing an external power connector, which I assume is more relevant for 1x cards who can only have 15W of 12V and 10W of 3.3V, there's little or no reason to be using it today.
The connector is not the problem. Anybody focusing on just the connector here is out in the weeds in terms of understanding.
The problem is the amount of current Nvidia is trying to shove through it without active load balancing.
All of this is regulated by Ohm's law. The voltage drop across a resistor is its resistance times the current going through it. The power loss in the resistor is the voltage across it times the current going through it. These relationships are always true. The only way to truly reduce the connector power loss is to lower the contact resistance. Load balancing only makes staying inside the connector spec easier -- you are correct, it does potentially increase power loss in the connector, but at least the connector won't burn up.
Generally speaking, connector contacts are rated on amperage. The manufacturer of the contact specs this current based on engineering knowledge of the expected resistance and thermal limits the connector can reasonably accommodate over its service life. It's done this way for convenience as you could also spec based on power loss ability. The Ohm's law thingy means both of these approaches are the same.
You care about the max current going through the pin, you generally don't care about the power loss. If you are within current limits, you are also, by definition, within the power limits of the connector - that extra watt is trivial.
As a side note, what is really aggravating this problem is 12V power voltage. One could go to 48V, and still be considered low voltage. Look what happens:
600W @ 12V = 50A.
600W @ 48V = 12.5A <-- 4x less
The PCIe 8 pin connector is rated for 150W @ 12V -> 150/12 = 12.5A. You could run your 5090 at 600W with a single PCIe 8 pin connector if you ran it at 48V.*
*A caveat here is the power supply design of the video card would be harder as you have to get that 48V down to the core voltage of the GPU. The bigger the voltage drop, the harder this is.
- Joined
- Dec 24, 2008
- Messages
- 2,269
how does current cause a connector to burn up? power loss in the connector IS what causes it to burn up. combined with possible positive feedback.
edit: forgot to mention i dont necessarily have an issue with the connector itself, just that nvidia used it on the 5090. although i did have bad experience with microfit 3.0 connectors from my last company.
- Joined
- Dec 24, 2008
- Messages
- 2,269
I'm trying to find some example, but currently failing. Datasheets for connectors that are more detailed typically dont just have a max current spec, they have a current allowed vs temperature chart, or some other plot that is temperature related. What im getting at is that's it not a simple "oh, im under 10amps, everything must be ok". It's more nuanced.
I've seen that too,it seems like the spec math doesn't quite balance in the register. I remember those AMD cards too.
There is a lot of engineering judgement in this too and I don't understand all of it. This whole power system stuff is a gigantic topic.
The Molex connectors used for PCIe 6 pin and 8 pin are actually rated higher by Molex than what the PCIe spec allows. Molex likely has a conservative rating, then the PCI people were more conservative than that.
Then there is the difference between the PCIe 6 and 8 connectors. Take PCIe 6, rated for 6.25A, add two ground connections, and poof, your good to go for 12.5A. Still only 3 power wires. Wut?
Another observed bit of engineering: 1080TI FE cards have an 8 pin (12.5A) and 6 pin (6.25A) connectors. The EVGA 1080TI FTW3 cards have two 8 pin connectors, so 12.5A each. I drive the second 8 pin connector off a 6 pin output of the power supply going through an adapter. At 100% TDP, I've never seen the second 8 pin connector draw over 6.25A. It's like someone at EVGA thought someone may plug a 6 pin output of their power supply into this through an adapter. If you push the card harder, the second connector will draw more than 6.25A, and it my case, it will click off the power supply, but it allows the 1080TI FTW3 to work at stock power levels with a smaller power supply.
That and a quiet two slot cooler are other reasons I miss EVGA...
I think the 1 number max in that case is the max current at the highest specified temperature, since that temp is probably driven by not melting the plastic.
Power loss in the connector is what causes it to burn up. Current times resistance is where the heat comes from. Is it dissipated into the surrounding environment. If the current is too high, the connector's temp will exceed what it can tolerate. If you keep the current within the specs for the connector, the connector will not exceed it's max value. The only way to stop the heat is either have zero current or zero resistance and neither of these operating points is very useful. So have as little resistance as possible and govern the current level and the connector will tolerate the heat.
Shit, I have one of those in the current design we are evaluating. I remember when we picked it, it was 1/2 the size of the connector it was replacing. Was pretty skeptical about the specs at the time, but we way underrated it in our application.
I do wonder who is really at fault here. Did the 12VHPWR connector manufacture overspec the design and then Nvidia took advantage of it by simplifying the board? With the value of these boards, it is not a risk I would have taken.
- Joined
- Dec 24, 2008
- Messages
- 2,269
I think we're almost in agreement here. So follow my logic:
1) Power loss in the connector is what burns it up
2) Every scenario I could come up with current balancing dissipates more heat in the connector than not current balancing
(and if you look at per pin max it can actually be A LOT higher)
3) We already suspect the connector, as it is used in the 5090, is marginal (again, not the connector itself, the connector as used to power 600 watts)
4) ANY increase in the connector power loss will make the problem worse. Current balancing will not fix the actual issue.
The main issue i had was with microfit 3.0 plugs that were 2x5 or wider ( like 12VHPWR for example
, but no we didn't actually use 12VHPWR, just generic microfit 3.0) . The very edge female receptacles could easily loosen up after multiple connections if you didn't connect/disconnect carefully. Generally 2x4 or smaller seemed fine. But this is anecdotal, so take it with a grain of salt.
Last edited:
1) Agree
2) Sure
3) Who's we? I suspect the PCB design.
4) False
Current balancing will only slightly increase the power lost in the connector, but it can completely solve the problem of the connector burning up. We know the latter is true because Nvidia themselves have been doing it successfully for decades.
- Joined
- Mar 18, 2010
- Messages
- 14,483
The point is to be able to catch a failure before damage occurs. The common bus design allows for an undetectable catastrophic cascading failure mode and requires the use of external current monitors to catch a failure. Separated pins can have a voltmeter on each pin on the GPU monitoring for any significant discrepancies, which can then trigger a software alert as well has throttle the GPU.
Reducing failure rates means a more robust design with more margin, no question about that. But if they want to run at the limit regularly, a system should have been implemented to catch failures early. Instead, this design hides failures until it's too late.
- Joined
- Dec 24, 2008
- Messages
- 2,269
Yeah. I agree. Im just more interested in the more robust design part. So that failure simply occurs much less often, and so there is less need for careful monitoring.
Just my two cents, but 5080s, 5070 TIs and 9070XTs (and maybe a 5070 as well) have also had melted 16 pin connectors - not just 5090s.
conservative engineering means it can easily take it. The bigger pins in the 6/8pin PCIe and 4/8 pin mobo connectors can take significantly more current before melting than the smaller ones in the new 16 pin one so there's plenty of margin in the connector. I suspect the main driving consideration for just using the extra pins as ground/sense though is that it simplified back compatibility. A new 8 pin cable carried exactly the same current as an old 2x 6 pin one; and it meant that a 2x6 to 1x8 adapter cable would be naturally load balanced.
- Joined
- Dec 24, 2008
- Messages
- 2,269
Oh, I thought we were in agreement:
Maybe I misunderstood you.
I can't remember, from memory, any of the designs from my previous companies that used current balancing. I didn't do the actual board design, so maybe im misremembering? 2.4kw PSU connected to single backplane, common 12V. From there connectors to 4 server nodes, each one using something that has multiple 12v, grounds, and all of the signals. Single 12V net on the server node. Other motherboards we would use 2x4 microfit 3.0, connect all of the 12v together. In Current company the engineer that is redesigning some of the power is annoyed that someone tried to load balancing on one of our designs (and failed miserably to implement it correctly) instead of just using MORE parallel connections, or more robust connector.
Do we know that? We can't use examples that didn't use 12vhpwr for comparison, because they had a lot headroom anyways. What's the failure rate on 3090s compared to 4090/5090? Do we have actual numbers? And is 3090 similar power consumption to 4090/5090?
Last edited:
- Joined
- Dec 24, 2008
- Messages
- 2,269
Agreed, that's where it would be interesting to see failure rates among these models. Without failure rates it doesn't tell me much that some 5070ti melted the 16pin connector.
- Joined
- Dec 2, 2004
- Messages
- 32,984
Yep, which proves it isn't the current that's the issue. It's just a rare defect.
- Joined
- Dec 24, 2008
- Messages
- 2,269
It's rarely a single thing that is the issue. Small defects can cause long term issues when running at high currents, for example.
On the first part, I've never said the power loss in the connector was zero. I also agreed with you that load balancing will increase the power loss in the connector - a very astute observation by the way. What I've been saying is the load balancing will manage the power loss in the connector to reasonable levels.
On that last point, of course we know it. Look at the power design of the 1080TI. Doesn't matter if the 1080TI isn't latest, watts are watts. Look at Buildzoid's analysis of the 1080TI boards.
On that middle point, I'm way to tired to understand it after a long work day, but it does look interesting. (This place doesn't seem to have a cheers emoticon). Will say that with small wires, putting more in parallel doesn't work that well, see my earlier posts. Big fat single works better than several in parallel without load balancing.
Without seeing the PCB layout it's hard to say. People seem to be getting all wrapped around the axle about how rare or not rare this is. On the 5090 it is a straight up messed up PCB design. Nvidia has already done designs that avoid this issue.
Boy, ain't that the truth. The worst fires I've put out have always been from multiple causes. It takes a while to realize multiple causes, then you have to figure out how to sort them, then bound them, and then hopefully having a plan to fix the mess.
I don't view this issue as one of those. If the new connector was rated the same as the previous connectors, and Nvidia had implemented load balancing, and this happened it would different. To me, this one just looks like a classic cutting costs too much and getting burned.
One thing that could help is more data points beyond just user reports of failure, and one way to get that would be to take a deep dive into each of the board designs showing this problem. We need to look beyond the 4090 and 5090.
Good point.
I disagree, the highest wattage cards have clearly had the most failures even though they're the least common cards. It's also not at all surprising given the lower margin for error which just gets smaller at higher wattage.
I tend to think they should have gone with even less pins than the old PCIe connectors but made them more robust.
- Joined
- Dec 2, 2004
- Messages
- 32,984
Got some stats? How is it "clearly"? They're just the most sensationalized. I hardly see that many unique cases of any cards melting, considering they've sold millions of 4xxx and 5xxx units.
Of course it is the current. What else would it be, demons? Little green men? Current through the contact resistance = heat. No current, no heat. More current, more heat.
A logical statement based on the physics of the problem, statistics barely required. Also supported by reports of 4090/5090 failures on cards running at reduced power.
Moar power means moar current which means moar potential heating -> the highest failure rate will be the highest power cards. How far down that extends is anybody's guess, but all else things being the same, the highest failure rate will be with the most powerful cards. Knowing about the low end would tell us just how marginal the current connector/PCB power supply design is. Can they fail at 400W, 300W, or 200W?
Lies, damned lies, and statistics - the tools of those with weak arguments. arestavo's and and ebernanut's statements are supported by the thin statistics available. Wanting more statistics to argue about those statements indicates someone who wants to obfuscate or doesn't understand the issue. Your statement about current above, makes me think the latter.
One thing about the elephant in the room...
The 4090 and 5090 are power pigs, not in the sense of performance per watt, but in the sense of fitting in a standard PC case. At 360W, the 5080 isn't as bad, but still pretty high compared to historical norms. These cards are a long way from the ~250W cards that used to be the norm.
The 4090s and 5090s are clearly pushing boundaries, and when you push boundaries, stuff breaks and new normals are formed.
The 4090 and 5090 are power pigs, not in the sense of performance per watt, but in the sense of fitting in a standard PC case. At 360W, the 5080 isn't as bad, but still pretty high compared to historical norms. These cards are a long way from the ~250W cards that used to be the norm.
The 4090s and 5090s are clearly pushing boundaries, and when you push boundaries, stuff breaks and new normals are formed.
- Joined
- Dec 24, 2008
- Messages
- 2,269
I never assumed you said the power loss is zero, my mistake if i somehow implied that. i dont know what this means: "load balancing will manage the power loss in the connector to reasonable levels." current balancing increases power loss in all of the scenarios im looking at. not only that, but the power loss is distributed more poorly (higher per-pin peaks). here are 3 more examples where i increased the poor contact resistances to 100mohm to show how it gets worse and worse . check out those per-pin waste heat values in balanced vs non balanced.
1080ti used a different connector. margin was higher.
I have a feeling, not backed by hard evidence (of course), that my 5090's connector with imbalanced pins didn't melt (prior to me finding the problem with a WireView 2 Pro) in large part because I've got four 180mm fans blowing in from the side. Airflow from the main opening side of the case is getting quite rare these days, with case manufacturers focusing on the looks of acrylic and glass sides. I don't doubt that the imbalance would have become worse with time and heating/cooling cycles, and would have eventually melted.
Pure speculation on my part since I threw that OEM adapter away and I wouldn't have used it to the point of failure anyways.
It means exactly what it says.
It looks like you are wrapped around the axle on the power loss distribution in the connector and load balancing making that "worse". It depends on how you define worse. It appears your definition is worse is more power dissipated in the connector. I don't care about that, I want to keep the power dissipated in the connector within the tolerance for the connector on every pin and I'm willing to pay for that with a little bit more power dissipation compared to a perfect connector without the load balancing.
The load balanced approach is much more tolerant of connector variation.
Engineering is all about compromise. As the philosopher Jagger stated "You can't always get what you want".
I have no idea how accurate the analysis above is without a datasheet for the connector. Is there a data sheet available for the connector? Your imbalance looks pretty high as it is an order of magnitude, 0.01 Ohms to 0.1 Ohms. Frankly, 0.1 Ohms is way too high. Just guessing, I would expect the variance to be something like +/-5%. Using 0.01 Ohms as nominal, that means 0.095 Ohms to 0.0105 Ohms. Do this and your power increase from load balancing is going to get a lot smaller. I even suspect 0.01 Ohms is too big. The datasheet would give you more accurate numbers to use.
Look beyond the connectors. The connector and PCB power supply input is a system. It works together and they are dependent on one another. If you only look at the connector you can't understand this issue.
A useless and trivial "solution" to this problem. Don't want the connector to burn up? Fine, don't plug it into a video card.
Blowing air across the connector will allow it tolerate more heat. Without a spec, it's hard to say how much more heat, but it will be higher than without the fan. The problem with the fan is how does the air get to the contacts inside the housing?
In the professional engineering world, having to put a fan on a mundane power connector would make you a laughing stock. UL also would not approve.
I haven't bought a case in quite some time and the modern ones are puzzling to me. Let's go from 250W cards to 600W cards and then shove them in a case that is thermally compromised for bling and coolness. Winning!
Last edited:
- Joined
- Dec 24, 2008
- Messages
- 2,269
Your end goal is current balancing. Im asking "why"? What do you want to achieve with balanced current? You want to force large amounts of current through connectors that have more resistance than other?
Is it purely the single datasheet spec you are trying to hit?
The resistance values are examples. I dont know the actual variation. I believe 10mohm is nominal. All the datasheets im finding are very sparse on any more detail than that (which gives me very little confidence in the datasheet).
You are trying to show me examples of current balancing that work better than non current balanced. But you've introduced an extra variable into your example that I can't isolate away. This example doesn't work.