[Falkentyne]

yeah, this is totally confusing, lol. I am getting 550W currently with this as it is ( PNY bios), close to 600W with the EVGA 500W bios, card is scoring high on the leaderboards, Not looking to tear down my loop honestly, too much work... As long as my card doesn't die, I am good, lol. Maybe I need to remove the 10mhom shunt on PCIE and replace with 5mohm? I dunno.

I'm not sure if PCIE is related here but it's possible.
Actually forget everything I typed previously.

Can you do me a favor and post your uncorrected GPU_Z values at maximum values for all wattage rails, please?
I Just looked at your TDP and TDP% and see they are wildly different. That means something you corrected is out of limits.

I Just looked at @olrdtg's post.
he seems to have the same type of power spread that you do
Except for that weird NVVDD1 input power (sum) that is showing up as way too low on yours.

RTX 3090 Founders Edition working shunt mod

Id recommend using HWINFO64 to monitor with. You can set up multiplyers for the power readings so you can easily see the true reading with out having to do the maths in your head. I'd absolutely LOVE to use HWINFO64, but sadly it does not like my motherboard sensors or something, as a minute...

www.overclock.net

I believe he's pulling 580W on his, so not that far from yours. I think he used 5 mOhm shunts also, unless he used 8 mOhm shunts.

Yeah there's something funky up with yours.
I Just ran a 400W power draw test and got 30 watts on that (42.9W corrected with 1.43x)
But you're at 16 watts at 500W, while olrdtg is at 40W at 580. But I really do not know what I'm talking about anymore.

Can you kindly post your GPU-Z "maximum" values for all wattage values, please?

 

[motivman]

will run a quick port royal test and post my values uncorrected for gpu-z and hwinfo. If you look at the screenshot I posted of the other member (thanh Nguyen), you will see he is also pulling 575watts (he shunted with all 5mohm) and his max NVVDD1 input power (sum) is 17.148W. both our cards are PNY 3090 with stock bios.

 

[olrdtg]

he seems to have the same type of power spread that you do
Except for that weird NVVDD1 input power (sum) that is showing up as way too low on yours.

I believe he's pulling 580W on his, so not that far from yours. I think he used 5 mOhm shunts also, unless he used 8 mOhm shunts.

Yeah there's something funky up with yours.
I Just ran a 400W power draw test and got 30 watts on that (42.9W corrected with 1.43x)
But you're at 16 watts at 500W, while olrdtg is at 40W at 580. But I really do not know what I'm talking about anymore.

Can you kindly post your GPU-Z "maximum" values for all wattage values, please?

I'm actually closer to 610 W under full load, back when I took that screenshot I messed up the scalar in HWInfo. I am using 3mOhm resistors. I had actually removed all of the original 5mOhm resistors from my GPU, and replaced them with 3mOhm rather than stacking for a permanent shunt mod.

As far as the strangeness some people are experiencing with power readings, I am starting to think it might be the result of the card itself and the power monitoring, since the card is trying to keep everything in sync, perhaps a shunt mod screws it up to the point where it no longer just reports incorrect power, but also messes up power reporting completely, as I've seen people with all 6 resistors properly modded reporting some strange values.

 

[Falkentyne]

I'm actually closer to 610 W under full load, back when I took that screenshot I messed up the scalar in HWInfo. I am using 3mOhm resistors. I had actually removed all of the original 5mOhm resistors from my GPU, and replaced them with 3mOhm rather than stacking for a permanent shunt mod.

As far as the strangeness some people are experiencing with power readings, I am starting to think it might be the result of the card itself and the power monitoring, since the card is trying to keep everything in sync, perhaps a shunt mod screws it up to the point where it no longer just reports incorrect power, but also messes up power reporting completely, as I've seen people with all 6 resistors properly modded reporting some strange values.

Yeah I noticed something.
Yesterday, I realized that my "GPU Chip power" was getting right on the wrong side of 300W, and I remember you said (I think it was you), that one power limit flag is chip power=300W.
So I decided to take apart the card again, remove the 5 mOhm shunt (snapped off easily) I had attached to it with the 842AR paint, and just go liberal on the paint and cover the dried cured paint that was on the edges of the original shunt and covered everything up, as I suspect the resistance was higher than what was on the 8 pins.

Reassembled and saw that GPU Chip power was now reporting lower than before, by about 20W. Keeping it more in balance with the 8 pins. Success?

But then I ran COD MW Ground War, at my multi-hours stable 165 mhz Core overclock....

and...
"DEV 6068 error".

Then I tried 150 mhz overclock which I have run for a week straight with zero crashes...
"DEV 6068 error".

Then I tried 135 mhz overclock.
"DEV 6068 error" after awhile of playing.

Then I tried 120 mhz overclock.....
No crashes.

Mind you, this was not due to temps at all. Temps were the exact same.

I'm starting to think that "GPU chip power" is related NOT To the V/F curve you see on MSI Afterburner, but to "Uncore" (NVVDD2) voltage. If the GPU chip power is higher, then it's possible one of the voltages will be fed higher to it to keep it stable, because it's (the boost logic, whatever) expecting the GPU Core frequency to also be higher. That's pretty much the only explanation I can figure out here. If GPU chip power is lower, then uncore voltage may be lower, which means a given core overclock will not be stable anymore. Of course I didn't bother trying +150 with 500 mhz memory instead of 600 mhz because I just wanted to kill people with my Legendary Malphas AS-VAL I Just unlocked and not crash repeatedly...

Sort of makes sense because now I'm back at the same overclocks I was using after installing the Seasonic 12 pin cable but before I started shunt modding.

 

[motivman]

These are my values after a port royal and timespy run. No longer hitting limit due to temperature (raised my fan RPN to 1500).... Honestly, I do not think we have figured this out completely yet... I feel more resistors might have to be shunted to completely remove power limit on these cards.

 

[Falkentyne]

These are my values after a port royal and timespy run. No longer hitting limit due to temperature (raised my fan RPN to 1500).... Honestly, I do not think we have figured this out completely yet... I feel more resistors might have to be shunted to completely remove power limit on these cards.

View attachment 2466364

Question.
Did your core clock or voltage drop during the test?
Go to MSI Afterburner and "lock" the voltage at 1.10v by pressing control F, then "L" on the voltage point.
This will lock the card at 3d clocks at the voltage the boost decides to use.
Then "clear" the logs in gpu-z and hwinfo64 and run your benchmark again.

 

[olrdtg]

These are my values after a port royal and timespy run. No longer hitting limit due to temperature (raised my fan RPN to 1500).... Honestly, I do not think we have figured this out completely yet... I feel more resistors might have to be shunted to completely remove power limit on these cards.

I've actually shunted all of the smaller 5mohm resistors on the FE card, and it made no difference on the power limit. If you shunted all 6 of the larger 5mOhm resistors, that should do it. I think something else is going on here.

 

[Falkentyne]

I've actually shunted all of the smaller 5mohm resistors on the FE card, and it made no difference on the power limit. If you shunted all 6 of the larger 5mOhm resistors, that should do it. I think something else is going on here.

What card does he have? Palit or Gigabyte? I forgot.
Did you see what I said about what happened when I lowered the GPU Core reporting a bit more, to my overclock stability?

SRAM seems "linked" to GPU Vcore but the curves seem different.
I wonder if it changes based on GPU core draw....

 

[olrdtg]

What card does he have? Palit or Gigabyte? I forgot.
Did you see what I said about what happened when I lowered the GPU Core reporting a bit more, to my overclock stability?

SRAM seems "linked" to GPU Vcore but the curves seem different.
I wonder if it changes based on GPU core draw....

He's got the PNY 3090.

Yeah I saw that, and it's rather interesting. It wouldn't surprise me if NVIDIA made some changes to GPU boost to more tightly control both memory and core clocks and voltages. Since the 3000 series is extremely power hungry, I'd be willing to bet they did whatever weird trickery they could to lower power consumption. It's entirely possible the two are somehow linked or the system tries to keep them balanced.

 

[Falkentyne]

He's got the PNY 3090.

Yeah I saw that, and it's rather interesting. It wouldn't surprise me if NVIDIA made some changes to GPU boost to more tightly control both memory and core clocks and voltages. Since the 3000 series is extremely power hungry, I'd be willing to bet they did whatever weird trickery they could to lower power consumption. It's entirely possible the two are somehow linked or the system tries to keep them balanced.

Sram is "Uncore" (NVVDD2) though. I don't believe memory voltage can be changed outside of Kingpin or i2c / Elmor tool.

0% voltage. (They were at 1.063v initially)

30% voltage.

100% voltage (they both started out at 1.10v initially, as I started at 100% first).

At one point, at 0%, Sram was at 1.063v and Core was at 1.081v...

Note: all wattage values in Hwinfo64 are multiplied by "1.41x".

 

[motivman]

something weird is going on with 2X8 pin cards and FE cards as far as power limit goes. Everyone I have seen reporting that they are still power limited after shunting are either on 2X8 pin cards or FE cards. With my PNY with stock bios, 500-525W is the max I can get the card to pull no matter what. when I shunted initially with 15mohm on all resistors, 500-513W was the max I could pull. Now with 5mohm, and 10mohms on the PCIE slot, theoretically, I should be able to pull 708w on the stock bios, yet my absolute max is 525W. when I flashed the EVGA 500W bios, my card is able to pull about 600W max... interesting.

 

[Falkentyne]

something weird is going on with 2X8 pin cards and FE cards as far as power limit goes. Everyone I have seen reporting that they are still power limited after shunting are either on 2X8 pin cards or FE cards. With my PNY with stock bios, 500-525W is the max I can get the card to pull no matter what. when I shunted initially with 15mohm on all resistors, 500-513W was the max I could pull. Now with 5mohm, and 10mohms on the PCIE slot, theoretically, I should be able to pull 708w on the stock bios, yet my absolute max is 525W. when I flashed the EVGA 500W bios, my card is able to pull about 600W max... interesting.

View attachment 2466374

View attachment 2466379

My FE just poured out 550 watts (before taking this screenshot in a second attempt, as the i can't seem to capture the OSD) trying to run Overwatch in 4k (1920x1080 + 200% render scale) with uncapped FPS. Power limit was at 109%, and Limit and Normalized were 1% from each other.
I was being limited by Total Board Power (400W), since Chip Power was reporting around 270W after the mod.

At 550W, TDP reached 111 or 113% or something and temps were getting way out of control so I stopped, saw this post and then decided to try again.

For some reason I couldn't capture the RTSS OSD with screenshot. If I took a SS, the OSD wasn't even there (I have no problem like that in Call of Duty).
But I had to stop the test at 74C.

You can see I was reaching 538W and barely starting to throttle. (throttling starts lightly when you get close to the power limit, then gets extremely tryhard when you reach the power limit). I was being limited by total board power.

The clocks remained at 2070-2055 mhz during the entire test. At 530W, a few times, voltage dropped down to 1.038v a "couple" of times then went back up to 1.055v, the beginning of minor throttling.

I saw anethema's post.
He is being limited by MVDCC. When he didn't shunt his PCIE, his MVDCC was like 230W and throttling. When he did, it dropped but is still extremely high.

Notice how low mine is? Remember I have 1.43x for all watt values entered.

 

[motivman]

My FE just poured out 550 watts trying to run Overwatch in 4k (1920x1080 + 200% render scale) with uncapped FPS. Power limit was at 109%, and Limit and Normalized were 1% from each other.
I was being limited by Total Board Power (400W), since Chip Power was reporting around 270W after the mod.

For some reason I couldn't capture the RTSS OSD with screenshot. If I took a SS, the OSD wasn't even there (I have no problem like that in Call of Duty).
But I had to stop the test at 74C.

You can see I was reaching 538W and barely starting to throttle. (throttling starts lightly when you get close to the power limit, then gets extremely tryhard when you reach the power limit). I was being limited by total board power.

The clocks remained at 2070-2055 mhz during the entire test. At 530W, a few times, voltage dropped down to 1.038v a "couple" of times then went back up to 1.055v, the beginning of minor throttling.

View attachment 2466381

I was just saying averages. I saw around 550W today playing control maxed out in 4k with no DLSS, and 4X AA, so it can achieve it, but its difficult. I will try to play control again, and post a screenshot. But remember, I am on a bios that maxes out at 366W, your FE bios is at 400W max if I am not mistaken.

 

[Falkentyne]

I was just saying averages. I saw around 550W today playing control maxed out in 4k with no DLSS, and 4X AA, so it can achieve it, but its difficult. I will try to play control again, and post a screenshot. But remember, I am on a bios that maxes out at 366W, your FE bios is at 400W max if I am not mistaken.

Yeah I'm pretty sure if I had soldered shunts, I could go above 600W, but I wouldn't be able to cool the damn thing on air cooling. The reason I know this is because my TDP% and TDP Normalized are next to each other, so I'm being limited by the TDP slider (at least in Overwatch 4k), and in Port Royal.

I Just did another run for you. The clocks didn't drop again until 74C, when I lost another 15 mhz when voltage dropped again from 1.081v to 1.075v because of high temps, but there was no actual throttling going on. You can see from the GPU-Z values, I'm being limited by the max 'stock' values (400W, 300W, unknown what max PCIE is, and 170W to the 8 pins).

edit
I just saw anethema's post after he modded slot. Do you notice he's actually geting a power limit warning again?

RTX 3090 Founders Edition working shunt mod

Yes idea would be to stack them. Should calculate that later in the tool....

www.overclock.net

Look carefully at his picture. What exactly on that chart do the two lines that go through the entire limit reasons rectangle, correspond to on his wattages?
The MVDCC. Nothing else peaked at that spot. Not PCIE, not anything else.

I also found a later post by him. Notice his TDP% and TDP Normalized?
And what is pouring tons of watts again?

[Official] NVIDIA RTX 3090 Owner's Club

Nice clocks on your FE on modern warfare. Have you tried to add more aa and af ? that might help under some circunstancias to increase load. Details are set to maximum. Nothing to change except adding render scale past 100%. It's a 1920x1080 144hz monitor overclocked to 165hz. I do get 100%...

www.overclock.net

And then he said he got thermal throttling when trying to mine...
so his problem is that MVDCC.

The reason mine is balanced is because MVDCC is so low, it's not limiting power. And I'm out of thermal headroom on the stock HSF anyway.

 

[motivman]

My FE just poured out 550 watts (before taking this screenshot in a second attempt, as the i can't seem to capture the OSD) trying to run Overwatch in 4k (1920x1080 + 200% render scale) with uncapped FPS. Power limit was at 109%, and Limit and Normalized were 1% from each other.
I was being limited by Total Board Power (400W), since Chip Power was reporting around 270W after the mod.

At 550W, TDP reached 111 or 113% or something and temps were getting way out of control so I stopped, saw this post and then decided to try again.

For some reason I couldn't capture the RTSS OSD with screenshot. If I took a SS, the OSD wasn't even there (I have no problem like that in Call of Duty).
But I had to stop the test at 74C.

You can see I was reaching 538W and barely starting to throttle. (throttling starts lightly when you get close to the power limit, then gets extremely tryhard when you reach the power limit). I was being limited by total board power.

The clocks remained at 2070-2055 mhz during the entire test. At 530W, a few times, voltage dropped down to 1.038v a "couple" of times then went back up to 1.055v, the beginning of minor throttling.

View attachment 2466381


I saw anethema's post.
He is being limited by MVDCC. When he didn't shunt his PCIE, his MVDCC was like 230W and throttling. When he did, it dropped but is still extremely high.

Notice how low mine is? Remember I have 1.43x for all watt values entered.

Just ran control again, pulled 536W with my card.... ultmately, I do not think we should compare behavior of FE card and reference card... they are on different bios, so will act a little differently. If I flashed the 500W evga bios, my card can easily pull 600W no problem, but I do not like that one of the 8 pins draws over 300W on its own, so that Is why I will stay on the PNY bios for 24/7, and only use the other bios for benching.

 

[Falkentyne]

Just ran control again, pulled 536W with my card.... ultmately, I do not think we should compare behavior of FE card and reference card... they are on different bios, so will act a little differently. If I flashed the 500W evga bios, my card can easily pull 600W no problem, but I do not like that one of the 8 pins draws over 300W on its own, so that Is why I will stay on the PNY bios for 24/7, and only use the other bios for benching.

View attachment 2466386

What was your GPU speed during the Control run? Did you monitor it?
That's important. I can't tell from your picture if your GPU voltage was throttling or not. Monitoring the core speed would tell us if the core is downclocking.

A heavy throttle on the core is like -200 mhz below initial clocks.

I think the fact that my PWR_SRC and MVDDC are maxing out at 43W and 23W during the 540W run, is why the actual power inputs and GPU core power (<300W) are limiting power rather than some strange rail.

When I applied paint to mod the PWR SRC and MVDDC, those were not too close to other parts so they were relatively easy to scrape, and I think I managed excellent contact, as well as PCIE slot. The 8 pins were hard, which is why I had problems originally with 8 pin #1 reaching 174W while 8 pin #2 was 135W, and then when I finally fixed that (after seeing that the 5 mOhm shunt I put on the 8 pins did nothing, and the 5 mOhm shunt I put on GPU Core (which I didn't even need to do) was acting like a 20 mOhm shunt with the paint, I removed the GPU core shunt and just painted over the remaining paint very well.

The entire issue with the Founder's Edition cards are the shunts not being flat. Edges are lower than the middle, which creates contact consistency problems. Stacked shunts are supposed to touch each other and have strong contact. I think that's why @olrdtg had such good results soldering on new shunts directly (but other people had problems removing the original shunts).

 

[motivman]

What was your GPU speed during the Control run? Did you monitor it?
That's important. I can't tell from your picture if your GPU voltage was throttling or not. Monitoring the core speed would tell us if the core is downclocking.

A heavy throttle on the core is like -200 mhz below initial clocks.

2070mhz on average. I am curious what your clocks are when you run a game like control, BF5 or mass effect andromeda maxed out in 4K. On other games, my card stays around 2145-2175mhz, sometimes as high as 2190mhz. Quick Question... why are you using 1.44 as your multiplier? If you shunted with paint, shouldn't it be 1.33? So based on your last GPU-Z screenshot, your power draw is actually around 500W.

 

[Falkentyne]

2070mhz on average. I am curious what your clocks are when you run a game like control, BF5 or mass effect andromeda maxed out in 4K. On other games, my card stays around 2145-2175mhz, sometimes as high as 2190mhz. Quick Question... why are you using 1.44 as your multiplier? If you shunted with paint, shouldn't it be 1.33? So based on your last GPU-Z screenshot, your power draw is actually around 500W.

If the core didn't drop below 2055, you weren't throttling. (When the core drops, the voltage will also drop with it). With a permanent super thick throttle bar like that, I would have expected to see 1850 mhz.

It just means the power limit flag was set by something not related to GPU Core, Total Board power, or the 8 pins or PCI Express.

I could of course be wrong. I'm wrong a lot.
What was the highest core speed? was it 2130 mhz+ ?

Edit oops.

 

[motivman]

If the core didn't drop below 2055, you weren't throttling. (When the core drops, the voltage will also drop with it). With a permanent super thick throttle bar like that, I would have expected to see 1850 mhz.

It just means the power limit flag was set by something not related to GPU Core, Total Board power, or the 8 pins or PCI Express.

I could of course be wrong. I'm wrong a lot.
What was the highest core speed? was it 21300 mhz+ ?

not sure with the hell is up with it, but I am satisfied with my card's performance now, so not worried about it. Thanks for all your help.

 

[zeejc]

So after reading through most of this thread in preparation for my new 3090 FE tomorrow and plans for a shunt mod. Am i right in assuming the op's 3 mOhm shunt replacement on all shunts with the proper tools is probably my best bet?