[Falkentyne]

So it looks like I FINALLY fixed the power balancing. Got the TDP% to actually be calculated correctly and now shows up as TDP Normalized 104.6% and TDP 98.2%, looping Heaven.
Working a lot better than before, GPU Board Power is actually in good balance now, and I'm able to pull as much as 570W (estimated from my kill-a-watt) from the video card in Timespy.

I only hope it remains stable and doesn't start acting strange with time...

Port Royal run. notice the normalized is no longer at 114% (my power limit is at 114%). So a lot better than before.

So here is what I did.
I scraped the two 8 pin shunts and cleaned them. This took HOURS. And I had electrical tape (Super 33+) around the board so I wouldn't scratch anything with the mini screwdriver. Not only did scraping take hours, I then had to clean the paint residue on the PCB that got under the tape, since I had to use alcohol to loosen the paint (some of it was IMPOSSIBLE to scrape otherwise). Again--I've asked multiple people here which shunt is 8 pin #1 and which is 8 pin #2, but yet no one seems to know at all. (how can you guys not know? Some of you guys have probing equipment). Anyway...

After the shunts were cleaned and silver shiny, I put a nice glob of paint over each of the contacts on the depressed edges (it sucks that FE shunts have the edges lower than the housing middle section), then I stacked a 5 mOhm shunt on top of it and pressed down firmly. With my Super 33+ tape protecting the PCB, I felt a lot safer about that. Then after it dried a bit, I coated the top and side edges of the "silver" part of the shunt with the paint, and tried to bridge it with the bottom shunt. There was barely any room to work with and plenty of paint got on the Super 33+ tape in the process.

After reassembling, results were at first awful (I only gained 40W from 400W on my kill-a-watt) but over the next hour, this slowly rose to about 500W (+100W). It briefly got a little better from there (KAW briefly registered 720W total), but then seemed to get slightly worse then tone down at 500W total (690W wall). And Heaven was mildly power throttling down to 2055 mhz and 1.038v, which I didn't like.

So I had a big brain moment last night.
I asked...well....what if I coat and bridge the top of the NEW shunts? If the entire circuit is acting like a modified shunt as is, then coating it fully should drop the resistance even more! The exact same way that coating the original 5 mOhm shunt without stacking lowers the resistance, but to less effect of course. I mean, this IS a science, right? Why would it not work? Scientific rules can't be broken...

So I coated and applied another layer on the silver edges and bridged them through the middle firmly...

And so far, success. (Depends how long it lasts...we all know these mods seem to work for awhile and might suddenly have an 8 pin or PCIE suddenly skyrocket to 175W or 79W....)

The drop in TDP Normalized% is the huge giveaway here. Instead of it being at my TDP slider (due to power rebalancing from the high 8 pin #1), it's now a lot more level, and Total Board Power is also reporting a bit higher than before.

 

[motivman]

So it looks like I FINALLY fixed the power balancing. Got the TDP% to actually be calculated correctly and now shows up as TDP Normalized 104.6% and TDP 98.2%, looping Heaven.
Working a lot better than before, GPU Board Power is actually in good balance now, and I'm able to pull as much as 570W (estimated from my kill-a-watt) from the video card in Timespy.

View attachment 2467714

I only hope it remains stable and doesn't start acting strange with time...

what did you do differently? and can you post a screenshot of a PR or timespy extreme run?

 

[zeejc]

There is ABSOLUTELY no power throttling here! how didn't anyone notice this before? I think I am gonna stack additional 8mohms on my 3 resistors except the 2 8 pin resistors and pcie resistor and cross my fingers!!!!!

Wait so basically the thought here is that ignoring the pcie slot and 2 8 pins and just shunting the others are all that need to be done then???

 

[motivman]

Wait so basically the thought here is that ignoring the pcie slot and 2 8 pins and just shunting the others are all that need to be done then???

not exactly. the 2 8 pins and pcie slot also have to be shunted (dante`afk currently has 5 mohms stacked on the 2 8 pins and 8 mohms stacked on his pcie)
, but it looks like the other 3 resistors needs to be double shunted (needs lower overall resistance). So Currently I have 5mohms on all resistors and 10mohm on the pcie resistor. What I am gonna do is leave the 2 8 pins and pcie as is, and stack an additional 8 mohm on the 3 other resistors, essentially changing their resistance to 1.9 mohm. dante`afk current has 3mohms on those 3 resistors, so his resistance is 1.88 mohm, but I do not have any 3 mohms resistors handy to stack, so closest I can get to that value is double stacking with 5mohm and 8mohm. hopefully 1.9mohm is low enough to defeat the power limt.

 

[Falkentyne]

not exactly. the 2 8 pins and pcie slot also have to be shunted (dante`afk currently has 5 mohms stacked on the 2 8 pins and 8 mohms stacked on his pcie)
, but it looks like the other 3 resistors needs to be double shunted (needs lower overall resistance). So Currently I have 5mohms on all resistors and 10mohm on the pcie resistor. What I am gonna do is leave the 2 8 pins and pcie as is, and stack an additional 8 mohm on the 3 other resistors, essentially changing their resistance to 1.9 mohm. dante`afk current has 3mohms on those 3 resistors, so his resistance is 1.88 mohm, but I do not have any 3 mohms resistors handy to stack, so closest I can get to that value is double stacking with 5mohm and 8mohm. hopefully 1.9mohm is low enough to defeat the power limt.

There's definitely something odd going on with the power balancing. My chip power and PCIE are reporting nice and low because I have a super thick layer of paint on them. Also for the SRC also. But the SRC is extremely weird. The SRC seems to be influenced by the power draw from the other rails. Shunt the 8 pins really well and the SRC drops low. Do a poor job and the SRC ends up showing 150W+ at load. There's definitely some shenanigans at work.

I think it might be worth it for people to shunt the SRC with even lower resistance than the 8 pins, to see if it affects how the SRC operates (example, if stacking, 3 mOhms on the SRC, 5 mOhms on the others, 8 mOhms on PCIE Slot, or something).

And MVDDC draw seems to be related to shunting the PCIE, not just shunting itself....

 

[Falkentyne]

Wait so basically the thought here is that ignoring the pcie slot and 2 8 pins and just shunting the others are all that need to be done then???

This is interesting actually.
You can try shunting the SRC, Chip Power and MVDDC with a 3 to 5 mOhm shunt first, and leave the 8 pins and PCIE slot alone, and then see how that affects the power draw. Then shunt the PCIE Slot and check. Finally shunt the 8 pins at the end. For science, of course.

This obviously would require a lot of repasting and every time you take the card apart, you're doing a risk of breaking it. But doing all these steps would tell all of us exactly how the power rails influence each other, and if going yeet on the SRC shunt would have any benefit versus having the SRC the same as the others.

The problem is, the watts of the SRC seem -directly- related to the power draw from the other rails......

 

[jomama22]

Tbh, the people having issues seems to be those using the paint method. My guess is that the resistances are all over the place on each individual shunt. One may be 15 mohms, one may be 7, one may be 3, and it's making it difficult to track anything logical down.

That is the kind of variations you're gonna get using that method unfortunately. Those that have solder stacked or replaced don't seem to have nearly as many issues unless it's just a poor soldering job.

 

[Falkentyne]

Tbh, the people having issues seems to be those using the paint method. My guess is that the resistances are all over the place on each individual shunt. One may be 15 mohms, one may be 7, one may be 3, and it's making it difficult to track anything logical down.

That is the kind of variations you're gonna get using that method unfortunately. Those that have solder stacked or replaced don't seem to have nearly as many issues unless it's just a poor soldering job.

Several people have failed desoldering jobs because they can't melt the tryhard FE solder, even after practicing and having no problems on a test PCB.
And some places are extremely space challenged...

Also....
It's because the stupid FE shunts have edges lower than the center, which makes accessing them a pain in the butt when trying to paint without making a mess and actually getting contact on the edges (look at the 8 pin next to the 12 pin area!)

Most of the shunts you can get full access to without shorting a VRM out by accidentally getting paint on the PCB, by just applying Super 33+ tape or Kapton tape around them.
But one of them on the 3090 FE--one of the 8 pins--is almost completely blocked by a choke, making it very hard to get something on that silver connection by the edge facing the choke. I'm starting to think that's 8 pin #1....because that's always been the one giving me problems, but I still am unsure which is 8 pin #1 and #2.

Do you see how tiny that spot is?

On a less dreary note, at least there's enough clearance to at least get electrical super 33+ tape around all the tiny parts around both of those shunts (besides the choke itself being tricky).
There's a choke by the PCIE Slot shunt too but there's a bit more space there.

The 3080 FE, on the other hand, has virtually unrestricted access to the two 8 pin shunts...

But too bad the top shunt has some tiny SMD's around it that are impossible to get Super 33+ tape or Kapton tape around because the 12 pin is "conveniently" in the way, and the MVDDC (bottom shunt by the V cutout) has more tiny components between the shunt and the 5 pin connection. If you got paint residue there, that would be hell to clean off, a Q tip can't fit in there... you would need like a toothpick surrounded by cloth or tissue with alcohol to wipe it up and you wouldn't have any space to get any real traction on it...

It's a LOT easier to mod shunts that are fully flat and flush...

 

[dante`afk]

good find falk. I might try that too tomorrow on the pcie slot shunt...because...guess what......it worked the whole day and then all of the sudden the pcie slot is capping again??????? whyyy

 

[motivman]

I put 3mOhms everywhere but on the pcie slot shunt (8mOhm) and on the pcie power plugs (5mOhm)

Tried this on my Reference card, was an absolute disaster. Card was throttling all over the place, scores were lower, clocks that were previously stable were crashing, couldn't even complete PR on stock clocks. removed everything, and shunted everything with 5mohms, now back to scoring high on benchmarks...phew

 

[mattxx88]

Tried this on my Reference card, was an absolute disaster. Card was throttling all over the place, scores were lower, clocks that were previously stable were crashing, couldn't even complete PR on stock clocks. removed everything, and shunted everything with 5mohms, now back to scoring high on benchmarks...phew

You mean mixing shunts type or using a full 3mOhm shunt?
glad you solved anyway

 

[dante`afk]

Tried this on my Reference card, was an absolute disaster. Card was throttling all over the place, scores were lower, clocks that were previously stable were crashing, couldn't even complete PR on stock clocks. removed everything, and shunted everything with 5mohms, now back to scoring high on benchmarks...phew

View attachment 2467781

interesting, throttling to what? power?

 

[dante`afk]

I hope this was the last time I had to disassemble the card to fix the pcie shunt.


nower power perfcap during port royal!

 

[Falkentyne]

I hope this was the last time I had to disassemble the card to fix the pcie shunt.


nower power perfcap during port royal!

View attachment 2467827

What was wrong with the shunt? Bad contact because of the lower silver edges?

 

[dante`afk]

Tried this on my Reference card, was an absolute disaster. Card was throttling all over the place, scores were lower, clocks that were previously stable were crashing, couldn't even complete PR on stock clocks. removed everything, and shunted everything with 5mohms, now back to scoring high on benchmarks...phew

View attachment 2467781

did you also put a 5mOhm on the pcie slot shunt? let it draw 143w?

What was wrong with the shunt? Bad contact because of the lower silver edges?

it works for some time until it stops, so I have to redo the soldering, for whatever reason.

I've put a 4mOhm now on it, but I think I'll replace it for 8mOhm, scared.

 

[Falkentyne]

did you also put a 5mOhm on the pcie slot shunt? let it draw 143w?


it works for some time until it stops, so I have to redo the soldering, for whatever reason.

I've put a 4mOhm now on it, but I think I'll replace it for 8mOhm, scared.

My guess is that since the two shunts' conductive edges are not in contact, the solder has to make the connection because of the 0.5mm drop on the original shunt. Also I bet the choke next to the shunt doesn't help the space issue either.

So when the solder dries, it contracts (shrinks) right? Leaving eventually poor contact on the edge of the original shunt, causing the resistance to rise.

Especially tricky since you dont want the solder to get on the pcb.

The fix for this is to use kapton tape completely around the shunt! Super 33+ tape works for paint but not for hot solder so there you use kapton tape. The hard part is getting a tiny strip between the edge of the shunt and that choke, unless you just cover the choke and press down in the tiny space above (or below) it...

If you're going to change it to an 008 stack, I suggest buying the kapton tape first, so you can do a better job on the solder while protecting the PCB. Until then just turn the power limit down in MSI AB.

 

[Falkentyne]

did you also put a 5mOhm on the pcie slot shunt? let it draw 143w?


it works for some time until it stops, so I have to redo the soldering, for whatever reason.

I've put a 4mOhm now on it, but I think I'll replace it for 8mOhm, scared.

Try something like this.

3M Polyimide Tape, 92, Scotch, 1/4" X 36 Yards, High Temp. - Masking Tape - Amazon.com

3M Polyimide Tape, 92, Scotch, 1/4" X 36 Yards, High Temp. - Masking Tape - Amazon.com

www.amazon.com

This is some of the best Kapton tape you can buy. Cheap kapton tape sucks and doesn't stick to anything (except itself).
Then you should be able to solder without worrying about solder bridging the PCB.

@Sky3900 you should consider that also for the future.

 

[dante`afk]

Yea I'm using something similar, specified for up 275c.


I wonder what happened to my pcie slot and mvddc watt readings, I replaced the 4mOhm with a stacked 8mOhm but they're still super low lol.

the only difference i'm doing now is, i'm bridging them also on the top with solder, like you do with the paint.

 

[Falkentyne]

Yea I'm using something similar, specified for up 275c.


I wonder what happened to my pcie slot and mvddc watt readings, I replaced the 4mOhm with a stacked 8mOhm but they're still super low lol.

the only difference i'm doing now is, i'm bridging them also on the top with solder, like you do with the paint.

View attachment 2467855

What benchmark or test are you running? You're getting throttling there.
I suspect the PCIE Slot is reading too low.
Can you have hwinfo running next to GPU-Z and show your "TDP Normalized" max tabs and TDP% tabs there?

I've already concluded that PCIE Slot shunt is linked with continuity to MVDDC (despite being on opposite sides of the PCB), just like drunknfoo found on the evga forums (that's why your MVDCC goes down the lower the resistance of PCIE), and he also said that 8 pin #1 and #2 are linked with SRC, and 8 pin #3 is linked to SRC and GPU Chip Power.

Edit
Bridging it at the top with solder is probably a mistake. Solder has literally very low resistance, so you're driving the resistance of the shunt even lower by bridging the top. With paint, it works because the paint is about 15 mOhms, which can get lower in resistance the thicker it is, like shown here:

 

[jomama22]

Yea I'm using something similar, specified for up 275c.


I wonder what happened to my pcie slot and mvddc watt readings, I replaced the 4mOhm with a stacked 8mOhm but they're still super low lol.

the only difference i'm doing now is, i'm bridging them also on the top with solder, like you do with the paint.

View attachment 2467855

Yeah, don't bridge solder like that. You probably don't have much, if any, real resistance now on the pcie lol. Could take a multimeter and just go across the shunt and see what it reads.