Overclock.net banner

581 - 600 of 634 Posts

·
Facepalm
Joined
·
9,484 Posts
I don't know what u mean by right side or left side
Right click the pictures. Open them in a new window.
Then click the mouse to expand them to full size.
 

·
Registered
Joined
·
130 Posts
Ah ok I see them now. So I redid no.1 pretty well and the power draw on it increased dramatically. Not sure if this is normal or not. Did you get the same thing for ur 3090 fe?
 

·
Facepalm
Joined
·
9,484 Posts
Ah ok I see them now. So I redid no.1 pretty well and the power draw on it increased dramatically. Not sure if this is normal or not. Did you get the same thing for ur 3090 fe?
Increased?
No, mine decreased. Did you make sure you got paint properly on the edges? The edges are lower than the middle so its important to actually press down on the paint to make sure it gets contact on the edges. (That's why you need Super 33+ tape--to make sure you don't get paint on the PCB when doing that).

Also, check the SRC shunt. If the SRC isn't shunted properly, it will cause power balancing issues as well. I noticed your SRC was 105W. It should be somewhere around 70W :)

Do you have any pictures of your mod? If you can take a high res picture of the shunts, perhaps we can look to see what's up.

I'm still not perfectly happy with mine although it's good enough (can be anywhere between 12 to 18W difference on #1 and #2), so I ordered some 3M Polymide Kapton tape (for protecting the PCB) and will eventually try shunt stacking with solder some time in the future. I practiced on two random 100 mOhm shunts and got a perfect stack by using Rosin Flux.
 

·
Facepalm
Joined
·
9,484 Posts
Do you know what is power limiting me now?
Nothing.
That junior bar can be ignored. You can check in GPU-Z directly above it and notice that neither the voltage nor clock dropped.

Question: How did you fix your 8 pin #1?
Were my instructions and pictures correct?
How did you remove the old paint (if you did?)

Your rails are almost perfect, except one. The only rail left that is a bit high (Definitely higher than normal, as it usually won't exceed 60W on a stock 400W draw!) is MVDDC (that is the shunt below the two 8 pins, on the GPU core side (it's right next to that weird gold pin connector thing at the V cutout).

You can touch that one up a bit if you want to. MVDDC is a bit strange as it doesn't scale upwards directly on power limits but on resolution and VRAM clocks. So it's possible you won't reach it, but you can definitely improve it. MVDDC shunt has a strange continuity with the PCIE Slot shunt, so shunting MVDDC is important too.

Ideal MVDDC for paint is 40-60W (or lower) and for stacked shunts (5 mOhm) is about 35W (or lower).
Default VRAM limit is 93.7W and max is 107.8W.

So give that a try if you have time!
How did you fix your 8 pin #1?
 

·
Registered
Joined
·
130 Posts
Nothing.
That junior bar can be ignored. You can check in GPU-Z directly above it and notice that neither the voltage nor clock dropped.

Question: How did you fix your 8 pin #1?
Were my instructions and pictures correct?
How did you remove the old paint (if you did?)

Your rails are almost perfect. The only rail left that is a bit high is MVDDC (that is the shunt below the two 8 pins)
You can touch that one up a bit if you want to.

Ideal MVDDC for paint is 40-60W (or lower) and for stacked shunts (5 mOhm) is about 35W (or lower)
I resoldered all my shunts with better flux. And yes the pic you sent did correspond to the 8pin no.1. I'm not sure why the MVDDC didn't change but I'm pretty sure the shunt was stacked properly
 

·
Facepalm
Joined
·
9,484 Posts
I resoldered all my shunts with better flux. And yes the pic you sent did correspond to the 8pin no.1. I'm not sure why the MVDDC didn't change but I'm pretty sure the shunt was stacked properly
Interesting. I thought you were using paint!
What flux did you use?

I'm guessing you may have not used enough solder around that shunt's edge, or you possibly (this is a guess) didn't apply flux on top of the first solder drop
(remember: 1) flux, 2) apply solder, 3) flux again. 4) put shunt on top of flux you added 5) heat to form a stacked solder shunt sandwich :)

Yeah the biggest problem with the FE cards (and Gigabyte) are the edges of the shunts being lower than the middle. So you need to flux properly and make sure the solder flows. Remember: when you apply flux and then solder on the ORIGINAL shunt, you also need to apply MORE FLUX on top of the solder you added, so that that flux 'attaches' to the stacked shunt you are adding, then the solder flows to it. That is how you insure a proper connection. And the solder blob on the edges of the original shunts should be higher than the middle housing as well (on shunts with flush edges to middle, it's MUCH easier to solder!!)

I practiced soldering two 100 mOhm shunts together using flux and it was pretty incredible just how the flux made the solder flow to itself. The solder will iterally wrap itself around the flux if you have flux connecting all the contact points.

I'm preparing to remove my paint and switch to solder on 8 pin #1 and #2 after I get my high temp Kapton tape. Eventually I'll also solder a shunt on GPU Chip Power but getting the paint off that is going to be very long and time consuming as there's a lot of it on Chip Power.

Another thing people miss when they say that they are applying a 15 mOhm shunt on PCIE and 5 mOhm on the other shunts (to "protect" the PCIE), is they are forgetting that there is continuity between the PCIE and MVDDC shunts. Not shunting PCIE properly or having PCIE higher resistance than the others will cause MVDDC to skyrocket also, causing throttle on both PCIE and MVDDC. On boards without fuses, it's just best to have all the shunts the same, then limit your power in MSI Afterburner.

Your PCIE seems fine so just look at MVDDC.
 

·
Facepalm
Joined
·
9,484 Posts
And what sort of Port Royal score do you get at 114% TDP slider?
 

·
Registered
Joined
·
315 Posts
After 1 month of shunted card (FE, 5mOhm soldered), still all fine and sweet values. i'm pretty happy with this solution

this is 3 hours CP2077 / Warzone gameplay

 

·
Facepalm
Joined
·
9,484 Posts
I removed the 842AR paint from 8 pin #1 and #2 and soldered on two 5 mOhm resistors. Yikes. What a 10 hour job of failing and almost destroying the video card by shorting the two tiny resistors left of 8 pin #1. Not a fun time getting that solder off. Then when I finally googled as to why the solder wasn't sticking to the 8 pin 1 shunt (8 pin #2 was done on the first try), they said I had to scrape like I never scraped before so I scraped until there was copper showing, then the solder actually stuck on the flux and I was able to finally get a connection. 10 hours later.

Still need to do GPU chip power. That has a ton of paint on it. Going to wait until my high temp polymide tape comes, then no risk of accidents after i spend like 2 hours scraping that paint off.

Whoever thought making a board with shunts with lower depressed conductive edges needs to have their little pink mars wrapped around a vat of molten lava.
 

·
High Clocker
Joined
·
3,659 Posts
I removed the 842AR paint from 8 pin #1 and #2 and soldered on two 5 mOhm resistors. Yikes. What a 10 hour job of failing and almost destroying the video card by shorting the two tiny resistors left of 8 pin #1. Not a fun time getting that solder off. Then when I finally googled as to why the solder wasn't sticking to the 8 pin 1 shunt (8 pin #2 was done on the first try), they said I had to scrape like I never scraped before so I scraped until there was copper showing, then the solder actually stuck on the flux and I was able to finally get a connection. 10 hours later.

Still need to do GPU chip power. That has a ton of paint on it. Going to wait until my high temp polymide tape comes, then no risk of accidents after i spend like 2 hours scraping that paint off.

Whoever thought making a board with shunts with lower depressed conductive edges needs to have their little pink mars wrapped around a vat of molten lava.

Its in there intrest to make it harder to mod.
Next gen I bet they are going to use shunts which are BGA style so you cant stack them. Then youll have to remove them and replace to change that power limit.
 
  • Rep+
Reactions: Falkentyne

·
Facepalm
Joined
·
9,484 Posts
@Falkentyne this is the best I can get it. Replaced the MVDDC shunt and its getting about the same value. Probably something to do with the power-balancing View attachment 2475649
I'll let you know after I remove the old MG 842AR paint from my "Chip Power" shunt and replace it with a 5 mOhm soldered shunt, probably tomorrow (if my back can handle the work). I'm waiting for the 3M Polymide 92 tape to arrive so I can tape over everything around that shunt so I don't accidentally short anything (you saw my adventures with accidentally solder shorting the two resistors by 8 pin #1, which was hell to get unshorted!!)

Right now my MVDDC is reading 0W at load, so I'll have to see what happens when I fix up Chip Power. I soldered on a 5 mOhm shunt on the two 8 pins, after removing all of the original 842AR paint, and now I'm being limited by GPU Chip Power at about 550-570W. (seems like 100% TDP is 230W limit, and 114% TDP is 285W).

I still have MVDDC shunted with MG842AR stacking a 5 mOhm shunt using the paint, so I'm not going to disturb that. SRC and PCIE Slot are paint only (and I'd prefer to keep it that way...scraping old paint is too much hard work). But it's going to be interesting what happens when I get chip power much lower. In previous tests, dropping Chip Power resistance causes SRC power draw to rise, but SRC also drops the better the two 8 pins are shunted also....it's weird.

Here's a quick run with Port Royal to show you where my Chip Power is. Very close to the throttle point. 1.9x used for board power multiplier with the stacked 5 mOhm soldered shunts on the 8 pins, although I don't know how accurate that is since I'm not sure how the PCIE slot power factors into it with just paint on the Slot.

2475650
 

·
Registered
Joined
·
315 Posts
i think i can confirm the FE is limited to a +/- 600w max overall absorbment
i see in HWINFO under the voice "NVVDD" a max output of 320w
i got it stacked shunt with 5mohm, so we can caluclate a X2 Hwinfo values
 

·
Facepalm
Joined
·
9,484 Posts
i think i can confirm the FE is limited to a +/- 600w max overall absorbment
i see in HWINFO under the voice "NVVDD" a max output of 320w
i got it stacked shunt with 5mohm, so we can caluclate a X2 Hwinfo values
What NVVDD are you talking about?
 

·
Facepalm
Joined
·
9,484 Posts

·
Facepalm
Joined
·
9,484 Posts
what's your thought? with port royal i got a higer 350w NVVDD (what this NVDD exactly is?)

Thank you for the screenshot.

It's definitely a rail that does NOT respond to shunting.
We do know that its a "Sum" of something, but we don't know what it's a sum of or even if what's being summed is visible on sensors!
I'm guessing that the bottom GPU Core NVVDD Output Power (sum) is a sum of the last "GPU Core NVVD Output Power + GPU Sram Output Power".

At least that matches up with this pre-shunt superposition run from last November:

2476131


The only sum I know 100% certain already is that GPU Core (aka GPU Chip Power in GPU-Z) NVDDD Input power (sum) + Sram input power (sum)= GPU Core (NVVDD2) Input Power (sum).
Since the two input powers add up and one is sram input power, it's only logical that the output powers add up there also.
 
581 - 600 of 634 Posts
Top