Overclock.net banner

381 - 400 of 634 Posts

·
Registered
Joined
·
103 Posts
I'm a little confused on which ones you used on the card, did you just put 3mOhm on all 6?
Edit - what if I don't want the card to pull an insane amount of power, I would like to just increase it to maybe 500w, should I stack all of them or just some and which mohm would I use?
I stacked 4mOhm on all 6. The 3mOhm figure was just my estimation of how much resistance is added from the silver paint at the joints.

If you are looking for a 500w limit, I would bridge across the top of all 6 resistors using the MG Chemicals 842 silver paint (paint only, no stacked resistors). This will process will give you somewhere between 500w-525w max power limit if done correctly. There is more information on this further back in the thread.
 

·
Facepalm
Joined
·
9,484 Posts
I stacked 4mOhm on all 6. The 3mOhm figure was just my estimation of how much resistance is added from the silver paint at the joints.

If you are looking for a 500w limit, I would bridge across the top of all 6 resistors using the MG Chemicals 842 silver paint (paint only, no stacked resistors). This will process will give you somewhere between 500w-525w max power limit if done correctly. There is more information on this further back in the thread.
Sky, do you have any idea why my power balance is like this?

I did some calculations, versus my Timespy scores I got on a STOCK card at 440W (114%), (pure stock, unopened, no changes in thermal pads/paste/shunts) with +100/+600, and I determined that to get the exact same scores (currently, post all mods and re-pads, repastes, etc), I need a 69% power target. Then, I took the GPU Board Power value shown in GPU-Z and in order to get 400W from the 69% run (which gave the same scores, about 21,015 Graphics score in Timespy), I need a 1.65x multiplier! 1.65x multiplier with paint? (GPU-Z reporting about 245W total board power for 12,015 graphics TS score). Okay....anyway....

Edit here is proof of that (although I don't have the original pre-mod runs, they are on my 3dmark history).
69% power target, +90 core, +500 memory.

2467413


Compare this to my "best" full run TS score before I got the paint (I got the paint on November 4th. this run was on October 25th.
21,022 graphics. +100/+600, 400 watts (114% TDP)

The absolute maximum graphics score I got (custom run, graphics test only) pre-mod was this one: 21082.


---------------------------

Take alook at this now.
This is my timespy and port royal run from this morning.
Note that I got these same scores (actually slightly better TS) when the mod was first done, back on november 11th, when the 8 pins reported balanced power draw.
Although here i'm at +150/+500, and was then at +150/+600.

2467405


2467406


So anyway...
My scores are the same, and I did a "Heaven" 1080p run last night and it seems I can reach 535W of power before hitting a "power limit" with the power slider at 114%. However the difference now is that the card apparently isnt "maxed out" and is hitting this "Unthrottled" limit, only if "Normalized" TDP is under 114%, while the total TDP is much lower (not even 100%).

It seems that "total TDP" only reaches 100% when I get close to 570W! But then some heavy throttling starts happening when 8 pin #1 gets to its power limit of 175W in GPU-Z...

2467407


Ignore the "throttle" and "Thermal" limit flags --you can see that "blip" on GPU-Z--this seems to be a bug in the latest Vulkan beta drivers. Doesn't happen at all in 456.98 Hotfix.

As far as those limits---here is a "High temp" 4K overwatch run from last night. Very high temps so clearly that power draw is pretty accurate.
2467408


You can see the throttle bar not being 'full size'--meaning the throttle warning flag was enabled because a power rail was approaching its limit. I actually saw up to 572W, but by this time chip reached around 80C so I stopped testing.

Anyway---back to the point:
What's going on with the 8 pin #1 and #2 and PCIE Slot readings here?
It's like the board seems to be 'normalizing" (TDP) itself based on close to 15 mOhms stacked (3.75 total shunt resistance=535W at 114% TDP), then showing throttle flags after that is exceeded.

@bmgjet said that the board is drawing heavily from 8 pin #1, because of me using new higher quality thermal pads. However why would it draw SO MUCH from 8 pin #1? That's a pretty huge balance difference.

Note that several days ago, I actually threw on a 5 mOhm shunt on top of the paint on the "Right side" shunt of the two 8 pins (I do NOT know which shunt is for 8 pin #1 or 8 pin #2... @olrdtg have you done your testing?) and it made NO difference in the power balancing! Then two days ago, when I got my Kryonaut Extreme paste, I threw a 5 mOhm shunt on top of the "other" (left side) 8 pin shunt! And still no difference in power draw balance!

So the power mod is working, the board seems to be drawing proper power, but reporting it in a VERY unbalanced way. And Overwatch being able to pull 570W on air....

Opinions?
 

·
Registered
Joined
·
861 Posts
So you are effectively using 1.5 mOhm × 5 and the pcie is 3 mOhm then correct?
hm, not sure. the pcie is 8mOhm + 5mOhm (because I think the original resistor is dead), so overall probalby only 8mOhm
3mOhm on everything else but the pcie plugs, where there are 5mOhms

Dante, in your above gpu-z shots, with the two "green power limit" bars, were you running Timespy (graphic test #2?)
yep.
 

·
Registered
Joined
·
103 Posts
Sky, do you have any idea why my power balance is like this?

I did some calculations, versus my Timespy scores I got on a STOCK card at 440W (114%), (pure stock, unopened, no changes in thermal pads/paste/shunts) with +100/+600, and I determined that to get the exact same scores (currently, post all mods and re-pads, repastes, etc), I need a 69% power target. Then, I took the GPU Board Power value shown in GPU-Z and in order to get 400W from the 69% run (which gave the same scores, about 21,015 Graphics score in Timespy), I need a 1.65x multiplier! 1.65x multiplier with paint? (GPU-Z reporting about 245W total board power for 12,015 graphics TS score). Okay....anyway....

Edit here is proof of that (although I don't have the original pre-mod runs, they are on my 3dmark history).
69% power target, +90 core, +500 memory.

View attachment 2467413

Compare this to my "best" full run TS score before I got the paint (I got the paint on November 4th. this run was on October 25th.
21,022 graphics. +100/+600, 400 watts (114% TDP)

The absolute maximum graphics score I got (custom run, graphics test only) pre-mod was this one: 21082.


---------------------------

Take alook at this now.
This is my timespy and port royal run from this morning.
Note that I got these same scores (actually slightly better TS) when the mod was first done, back on november 11th, when the 8 pins reported balanced power draw.
Although here i'm at +150/+500, and was then at +150/+600.

View attachment 2467405

View attachment 2467406

So anyway...
My scores are the same, and I did a "Heaven" 1080p run last night and it seems I can reach 535W of power before hitting a "power limit" with the power slider at 114%. However the difference now is that the card apparently isnt "maxed out" and is hitting this "Unthrottled" limit, only if "Normalized" TDP is under 114%, while the total TDP is much lower (not even 100%).

It seems that "total TDP" only reaches 100% when I get close to 570W! But then some heavy throttling starts happening when 8 pin #1 gets to its power limit of 175W in GPU-Z...

View attachment 2467407

Ignore the "throttle" and "Thermal" limit flags --you can see that "blip" on GPU-Z--this seems to be a bug in the latest Vulkan beta drivers. Doesn't happen at all in 456.98 Hotfix.

As far as those limits---here is a "High temp" 4K overwatch run from last night. Very high temps so clearly that power draw is pretty accurate.
View attachment 2467408

You can see the throttle bar not being 'full size'--meaning the throttle warning flag was enabled because a power rail was approaching its limit. I actually saw up to 572W, but by this time chip reached around 80C so I stopped testing.

Anyway---back to the point:
What's going on with the 8 pin #1 and #2 and PCIE Slot readings here?
It's like the board seems to be 'normalizing" (TDP) itself based on close to 15 mOhms stacked (3.75 total shunt resistance=535W at 114% TDP), then showing throttle flags after that is exceeded.

@bmgjet said that the board is drawing heavily from 8 pin #1, because of me using new higher quality thermal pads. However why would it draw SO MUCH from 8 pin #1? That's a pretty huge balance difference.

Note that several days ago, I actually threw on a 5 mOhm shunt on top of the paint on the "Right side" shunt of the two 8 pins (I do NOT know which shunt is for 8 pin #1 or 8 pin #2... @olrdtg have you done your testing?) and it made NO difference in the power balancing! Then two days ago, when I got my Kryonaut Extreme paste, I threw a 5 mOhm shunt on top of the "other" (left side) 8 pin shunt! And still no difference in power draw balance!

So the power mod is working, the board seems to be drawing proper power, but reporting it in a VERY unbalanced way. And Overwatch being able to pull 570W on air....

Opinions?
You definitely have a large discrepancy there. What the cause of 8-pin1 reading so much higher than 8-pin2 is, I'm not sure. MVDDC is also very low.

Here are some ideas for trouble shooting...

1. Take the card apart and inspect each modded shunt with a 10x magnifier for: good contact, uniform paint thickness/coverage, delamination, discoloration, any paint bridging across to other components or on other areas of the PCB where it shouldn't be. Re-work anything that doesn't look right.

Note: I remixed the paint in between each application on every shunt. I'm just guessing that the silver flake settles out pretty fast, so, I did this to try and get the most consistent results.

2. If you have a sensitive enough multimeter you could try measuring the resistances of each shunt to see which one is off and then re-work it.
3. I would try re-doing the mod on the two power connector shunts to see if that solves the problem. If you are stacking shunts use new ones in case one of them is bad.
4. If #3 doesn't work, maybe try removing all of the mods to see if you get proper power readings with a stock setup. At least then you know something on the card is not the problem.

Also worth noting that the fans draw off 8-pin1. It always reads a few watts higher than 8-pin2 unless the card is in 0 rpm mode.

Here's a screen shot of my card stats running the PR stress test just for comparison. Balls to the wall settings for 15 minutes. Gpu-z is showing the average value. I'm hitting a power limit well below max TDP at times, but, none of the power readings look to be hitting their max and everything looks pretty well balanced... Wondering if there are some other power limits which don't get reported that I am running up against. These results seem to vary by application to some degree as well, PR will draw 109% TDP max, but Metro will draw 113.5%. %TDP vs. Normalized TDP can also vary a lot by application. In furmark, for instance, I will get large persistent discrepancies between the two due to the high memory/memory controller load. Whereas 3dmark and games report pretty close to the same values most of the time.

2467501


2467503
 

·
Registered
Joined
·
28 Posts
I stacked 4mOhm on all 6. The 3mOhm figure was just my estimation of how much resistance is added from the silver paint at the joints.

If you are looking for a 500w limit, I would bridge across the top of all 6 resistors using the MG Chemicals 842 silver paint (paint only, no stacked resistors). This will process will give you somewhere between 500w-525w max power limit if done correctly. There is more information on this further back in the thread.

Thanks for the reply. Iread through the pages but I feel like I understand more of stacking/replacing rather than bridging now lol. Are you saying just scrape the r005 until you see the silver on the sides then basically put a layer from the pen connecting both sides creating a "bridge"?
 

·
Facepalm
Joined
·
9,484 Posts
You definitely have a large discrepancy there. What the cause of 8-pin1 reading so much higher than 8-pin2 is, I'm not sure. MVDDC is also very low.

Here are some ideas for trouble shooting...

1. Take the card apart and inspect each modded shunt with a 10x magnifier for: good contact, uniform paint thickness/coverage, delamination, discoloration, any paint bridging across to other components or on other areas of the PCB where it shouldn't be. Re-work anything that doesn't look right.

Note: I remixed the paint in between each application on every shunt. I'm just guessing that the silver flake settles out pretty fast, so, I did this to try and get the most consistent results.

2. If you have a sensitive enough multimeter you could try measuring the resistances of each shunt to see which one is off and then re-work it.
3. I would try re-doing the mod on the two power connector shunts to see if that solves the problem. If you are stacking shunts use new ones in case one of them is bad.
4. If #3 doesn't work, maybe try removing all of the mods to see if you get proper power readings with a stock setup. At least then you know something on the card is not the problem.

Also worth noting that the fans draw off 8-pin1. It always reads a few watts higher than 8-pin2 unless the card is in 0 rpm mode.

Here's a screen shot of my card stats running the PR stress test just for comparison. Balls to the wall settings for 15 minutes. Gpu-z is showing the average value. I'm hitting a power limit well below max TDP at times, but, none of the power readings look to be hitting their max and everything looks pretty well balanced... Wondering if there are some other power limits which don't get reported that I am running up against. These results seem to vary by application to some degree as well, PR will draw 109% TDP max, but Metro will draw 113.5%. %TDP vs. Normalized TDP can also vary a lot by application. In furmark, for instance, I will get large persistent discrepancies between the two due to the high memory/memory controller load. Whereas 3dmark and games report pretty close to the same values most of the time.

View attachment 2467501

View attachment 2467503
hi @Sky3900, thank you for the reply!! I know you're busy so I always appreciate your feedback and hard work.

The next time I tear down my card, I'll look at everything and double check. But I can already tell you there's absolutely lnothing shorting anything (this much I checked 100%).

However what I can tell you is that the shunt mod is at least working correctly and seems to be based on throttling at around a shunt resistance of "about" 3.68 mOhms or around 540W'ish at 114% slider. I used my Kill-a-watt and cross referenced and verified my power draw at my 535W profile (114% TDP) and my 400W profile (71% TDP). (I got the 71% value by comparing and equalizing Timespy tests I did before I ever did the mod, back when you were discussing your original success), with my saved 400W + 114% TDP Timespy runs, and then reducing the power limit slider until I got close to equal results. Once I did that, I got 275W GPU Board Power Draw for my 21,015 Graphics score, then multiplying that by 1.64* gave me 400W...thus that's my hwinfo multiplier, so 71% TDP slider for 400W.

Anyway a little while earlier, I hooked up my Kill a watt and set my board to 71% TDP and ran Heaven, which reported 398W (via throttling) in Hwinfo64, and 580W on the kill a watt.
Then I set the slider to 114% and ran Heaven and it got up to 522W (no throttling except the normal temp scaling), and 710W on the Kill-a-watt. 125W increase on kill-a-watt, 125W increase in hwinfo64. Matches right up. Temps were a toasty 76C :)

So far, the board seems to start throttling lightly between 540W-550W (Heaven doesn't get there) and heavily at 565W (I had Overwatch while running at 4k (1080p + 200% render scale) reach 570W at 79C-80C!!!!). And this "3.68 mOhms" shunt reporting seems to be based on 8 pin #1 reaching 168W (start of throttling) to 175W (heavy throttling), rather than "Total board power" reaching 400W in GPU-Z. What's really strange is, Total Board Power reaches almost 350W when 8 pin#1 reaches 175W. :)

Anyway I'll look at the balancing later to see if I can fix it sometime. (If it gets worse I'll do it sooner) but it seems to be stable.
 
  • Rep+
Reactions: Sky3900

·
Registered
Joined
·
103 Posts
Thanks for the reply. Iread through the pages but I feel like I understand more of stacking/replacing rather than bridging now lol. Are you saying just scrape the r005 until you see the silver on the sides then basically put a layer from the pen connecting both sides creating a "bridge"?
Yep, that's correct. Gently scrape the contacts with a mini flat blade screw driver until all the dull grey coloration is gone and the surface is shiny. Using q-tips, clean the shunts with IPA and dry. Then connect both contacts by painting over the top of the shunt creating the bridge. The resistance of the paint goes down as the layer thickness increases, so, make sure that the wet paint layer is bulging up a bit. It will lose much of it's volume when cured. Keep it mixed well and shake the pen in between each shunt application.

You'll probably want to replace the thermal pads as well. IMO, they don't reseat properly after disassembly. Although, you can get away with adding a layer of thermal paste in between the pads and SMTs. I did this on my last disassembly and it seems to be working well so far.
 

·
Registered
Joined
·
315 Posts
finally i got it done, shunted mine 3080 TUF.
i was runned out of flux but i cannot wait tomorrow to have the new one :D and went soldering without it, the work is not visual perfect but seems working. so don't judge my welding skillz :rolleyes:

3 mOhm on all resistors (pcie one is not pictured)




this was my situation before shunting



And this one after, all seems to be fine



today i will receive a 3090 FE also, i think i'll operate that too

i noticed with a wattmeter at the wal, my overall power consume went 40w higer at the same settings (0.925v @2130mhz, superposition 1080p extreme) 460w before, 490/495w after
 

·
Registered
Joined
·
13 Posts
Ok so this thread seems to be a never ending search for whats optimal or options someone has for the FE. I'm currently debating stacked 3 mOhm on everything with a single 3mOhm on the pcie. I Will be getting a block, liquid metal the core, and padding it up nicely (following some things in this thread)and putting a smal block (6 dimm bitspower)on the backplate as well. This should cover all my bases then right?
 

·
Registered
Joined
·
315 Posts
I have to make further tests, but even if gpuz shows nice and optimal resaults, it seems to me that the card is still locked and not fully "open"
anyway got to dig better, i need time and having little sons don't help 😒
tonight i put the new waterblock on too
 

·
Facepalm
Joined
·
9,484 Posts
I stacked 4mOhm on all 6. The 3mOhm figure was just my estimation of how much resistance is added from the silver paint at the joints.

If you are looking for a 500w limit, I would bridge across the top of all 6 resistors using the MG Chemicals 842 silver paint (paint only, no stacked resistors). This will process will give you somewhere between 500w-525w max power limit if done correctly. There is more information on this further back in the thread.
@Sky3900 if you're here please check your private message ASAP, it's really important :(
 

·
Registered
Joined
·
861 Posts
as of today my pcie is throttling again, this was during timespy run.

like wth. I didn't make any changes. why is my mvddc so abnormally high also

2467680
 

·
Facepalm
Joined
·
9,484 Posts
as of today my pcie is throttling again, this was during timespy run.

like ***. I didn't make any changes. why is my mvddc so abnormally high also

View attachment 2467680
Check the PCIE shunt.
MVDDC being high like that usually happens when PCIE is throttling.
 

·
Registered
Joined
·
861 Posts
Check the PCIE shunt.
MVDDC being high like that usually happens when PCIE is throttling.
getting tired of redoing the soldering all the time :D

Seems it dropped, even got rid of the power warning on test 2 of timespy, but now all of the sudden vOP and vREL, which I didnt have before.

2467681
 

·
Registered
Joined
·
315 Posts
as of today my pcie is throttling again, this was during timespy run.

like ***. I didn't make any changes. why is my mvddc so abnormally high also
dunno if it matters, but i see in my 3080tuf shunted, that even with a good shunt, gpu still throttle for PL
i think they have further power limiter hided somewhere
 

·
Registered
Joined
·
313 Posts
getting tired of redoing the soldering all the time :D

Seems it dropped, even got rid of the power warning on test 2 of timespy, but now all of the sudden vOP and vREL, which I didnt have before.

View attachment 2467681
woah, this is perfect! seems like you are no longer power limited anymore based on that screenshot. Please report what shunts you have stacked! You might have solved this dilemma!!!!!!
 

·
Facepalm
Joined
·
9,484 Posts
getting tired of redoing the soldering all the time :D

Seems it dropped, even got rid of the power warning on test 2 of timespy, but now all of the sudden vOP and vREL, which I didnt have before.

View attachment 2467681
VOP and VREL is what you want to have.
VOP means voltage has reached its maximum operating point on the V/F Curve via silicon hardware.
Voltage reliability means the boost clock has reached its maximum allowed clock for that voltage (via bios)

It's possible to have VREL without VOP. VOP simply means the maximum voltage "tier" point via hardware (silicon) has been reached. You can get VREL without being at the highest voltage tier.
(note: usually getting VOP also means your core clocks will increase by 15 mhz).
 
  • Rep+
Reactions: dante`afk

·
Registered
Joined
·
313 Posts
yea, no matter how lower I'd go with the resistors, something on the card will throttle anyway.

run port royal and you'll see full power throttling.

View attachment 2467330


here's the card with 3mOhm everywhere but on the pcie slot and 2 plugs (5mOhm's there).


Even though I have a corsair ax1200i. I don't think 3mOhms on the plugs is a good idea ^^
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!!!!!
 
381 - 400 of 634 Posts
Top