[Falkentyne]

Quick question since I am going to re-attempt my shunt mod this weekend (I forgot to scrape off the conformal coating enough last time), as I am seeing mixed advice on what resistance shunts I should be using for each stack. I plan on stacking 5 mOhms resistors on top of the existing ones on my reference PNY card.

Some posts say to use the same resistance for every shunt, whereas others say to use something higher (10 mOhms or so) for PCIE Slot Power.

Just wanted to confirm which was best, and if simply stacking 5 mOhm shunts on everything will be fine.

Cheers!

The PCIE Slot being higher than the others was just a safety precaution for those who have 10 amp fuses on their PCIE Slot.
If you have a 20 amp fuse or no fuse at all, just go yeet the slot and put it the same as everything else.

Soldering shunts with flux is really the most reliable way. Now that I am learning how to solder, I'm a solder fanboy. If you get a good connection by using proper flux, good solder (Kester 60/40 solder is really the best solder) and a good temperature regulated soldering iron (those starter 25W irons part of beginner kits) are fine for soldering mouse and keyboard microswitches into PCB holes, but they are atrocious for shunt mods because the PCB ends up soaking up all the heat--that's its job after all) and--most importantly, some high quality 3M Polymide High temp tape to protect your PCB, and a good set of tweezers, it's actually easier than paint mods. No guesswork whatsoever.

Plus, you can always grab some old dead video card PCB with unused solder joints and practice soldering shunts on until you become a pro at it.

 

[Falkentyne]

I stillhit power limit by set 1.1v at 2190mhz. Any reason? Stock PNY bios. I tried gigabyte but it is the same. Did stack 6 resistors like a guy above.
View attachment 2464224

I know this is an old post, but these power limits people are getting "around 550-600W" after doing a perfectly shunted board with soldering (not paint) seem to be related to hidden power rails in the BIOS, that the six (or 7) primary shunts do not control. They seem to be linked to MSVDD or NVVDD power rails, because one thing I noticed in my own testing is, if you set TDP to 100%, after a certain amount of "corrected" power--usually somewhere around 470W after you multiply by the mod multiplier, your "effective" clocks in HWinfo64 (or in Thermspy 3.01) drop far below your requested clocks (the normal clocks you see in MSI Afterburner or the regular field in HWinfo). For example, 2070 requested, 1970 effective, etc. This is then followed by actual throttling of the requested clocks as the TDP goes even higher.

We know it's related to MSVDD due to what happens when people try to modify the V/F curve and set the 1.1v voltage point too high compared to the previous voltage points, and the same thing happens. People with Kingpin cards and access to the classified tool were able to modify MSVDD which brought the effective clocks back up. So it's likely a power rail related to one of these two rails that either isn't shown in the Ampere Bios Editor, or in Hwinfo64. There are two small shunts (three actually, but one is connected to the RGB header), in a 1206 size package, 5 mOhm, next to the GPU Chip Power and PCIE Slot shunts on the backplate side. Rumor has it that these two shunts may affect these rails. While @olrdtg modded these, he used wire, which multiple experienced solderers (Elmor, Oldidey) say is terrible for milliohm shunts, because the wires add far too much resistance. Either me or @dante`afk will be properly solder stacking with 5 mOhm 1206 shunts very soon (not on the RGB 1206 shunt, unless the first two do nothing) and will report if this affects the power limits limiting these (and reference cards that are NOT a Strix), to 550-580W before tripping a high TDP Normalized %.. TDP Normalized % in HWinfo64 seems to see these limits and sees that the individual rails for these limits are much higher than the primary rail limits (with respect to TDP), which supports that theory, along with effective clocks being affected first (at 100% TDP).

Oddly enough, at 114% TDP, effective clocks don't seem to be affected at all. Instead, when the card gets to about 570W TDP, it just throttles. If I had to take a guess, it's the other rail hitting a limit first. The logical explanation for this would be if the "first" rail has a lower "default" (based on 100% TDP slider) value than the second, but a higher "maximum" value, while the second rail has a higher "default" but lower "maximum". But without a schematic of a FE card, this all heresay and blind guesses.

 

[TJD269]

Hey folks, trying to avoid a wall of text as I just would like some feedback from the electrical wizards that reside in this thread. I haven't used this website since 2014 when I was a freshman in college... how time has passed and how funny to see 780 Ti and Korean IPS owner clubs in my sig. Have been using reddit pretty extensively since for the computer hardware forums, but nice to be back.

Anyway, I have followed this thread as a lurker pretty thoroughly for the past couple months because I've had a 3090 FE I got from Best Buy in late October. For the most part, love the card, but she's got a bit of coil whine and in general clocks can be all over the place due to being super power limited, however I'm clearly preaching to the choir.

So, over the past couple months, I've also noticed my fan curve being overridden under certain workloads with no specific reasoning to 100% which even with my wireless Hyper X Cloud II's is audible and annoying. Further research shows, similar to this thread and others, that my memory junction temp is hitting a TJ max of 110c which is obviously stupid hot and sure enough is what's causing my fans to ramp to 100%.

Of course, this all meant it was time to take apart the GPU and re-thermal pad it as well as shunt-modding while I'm in there. Now I did the only silver paint method despite soldering my whole life as I was still too worried about the risk associated and the resistors are in a few tight spots I just deemed I couldn't do without being drenched in sweat after doing just one. I also mistakenly only got a single thermal grizzly thermal pad kit and not 3 like I'm supposed to (just reordered 3 packs and they come Thursday so I plan to redo this and likely add more silver paint if necessary).

So I only did the backplate and it was hardly enough but I made it work, just not to my expectation, but that's what I get for not reading I need 3. I taped around all my resistors, scraped the coating off of the ends of the shunts, and very carefully painted the actual resistor from end to end trying to make as much meaningful contact as possible. I also tried to make each one as consistent as the last obviously so we don't piss off Nvidia's power balancing.

Now, the results were interesting! I did this all at work and brought the card home and fired her up no issues. Under load, the memory junction temp was definitely better by a few degrees and didn't get as hot as fast, but it inevitably still gets to about 106 on a long gaming session or benchmark and that's what I get for not reading everything around buying 3 kits... Still an improvement and most importantly I can still use my card until then. The other thing is that is definitely helped despite me not being able to really gauge a power difference. Check this out:

Previous results with the 3090 maxed out in every way in my 9900KF build:

Port Royal: I scored 13 511 in Port Royal
Time Spy Extreme: I scored 9 334 in Time Spy Extreme

New results last night with my 5950x somewhat dialed in:

Port Royal: I scored 13 830 in Port Royal
Time Spy Extreme: I scored 10 775 in Time Spy Extreme

Obviously besides a higher score, but look at those sustained clocks!!! That's a significant improvement over my past average clock in both benches and just in general I could immediately tell it wasn't dipping down into the 1800MHz and 1900MHz range in some loads anymore which is great. This card literally would dip down as far as 1890 depending on the workload no matter what slider I maxed out. Now, the lowest I was seeing was 1980MHz and some points and that was super rare... also, lower peak clocks, but maybe that's a driver thing? Not sure, I was hitting 2200MHz before somehow granted it performed worse so maybe a moot point. I know HAGS for sure affects it, but I'm 90% sure I had it on in the first benches too.

What's weird to me is what GPU-Z is showing and I'm hoping I can squeeze just a hair more out of the card considering what it shows:

This screenshot was taken after one of my earlier runs yesterday before I maxed the voltage slider (which seems to do literally nothing this gen and especially for this card) and also before Windows's HAGS was turned on (which effectively lowers my clocks a bit but gives me a pretty hefty performance gain), but my power slider was maxed in MSI afterburner. Notice that pin 1 and 2 are slightly different and during some benches, they can be usually as far apart as 35w as well as pin 2 is almost always .1v higher than the other... clearly I knew these would change so I thought they'd still stay the same, but again, I definitely saw a performance improvement related to power so it's far from a negative. From what I remember at the very least is that "Board power draw" shows no difference from stock as it should be around 400w to begin with, but I figured this would actually show less despite truly drawing more power.

Anyway that's mostly what I've got. I ran an insane amount of benches last night if you need any more to look at, but I'm really hoping one of the shunt mod gurus in here can give me an idea of what I'm dealing with, what my numbers mean, why the pins are different, how I can push it a bit more (by re-shunt-modding), and maybe give me an idea of what I'm actually pulling power-wise now. I hooked up the kill-a-watt last night and saw all of a 400w jump from when the benchmark would load (2% GPU usage) to when it started and went to 100% usage, so it's definitely there if not pulling more. I also now know with my current CPU settings and the card maxed, I should see around 670w from the wall give or take so I can use this for reference if I redo the shunt mod to find any differences.

Thanks for reading and thank in advance for your time and help!

 

[Falkentyne]

Hey folks, trying to avoid a wall of text as I just would like some feedback from the electrical wizards that reside in this thread. I haven't used this website since 2014 when I was a freshman in college... how time has passed and how funny to see 780 Ti and Korean IPS owner clubs in my sig. Have been using reddit pretty extensively since for the computer hardware forums, but nice to be back.

Anyway, I have followed this thread as a lurker pretty thoroughly for the past couple months because I've had a 3090 FE I got from Best Buy in late October. For the most part, love the card, but she's got a bit of coil whine and in general clocks can be all over the place due to being super power limited, however I'm clearly preaching to the choir.

So, over the past couple months, I've also noticed my fan curve being overridden under certain workloads with no specific reasoning to 100% which even with my wireless Hyper X Cloud II's is audible and annoying. Further research shows, similar to this thread and others, that my memory junction temp is hitting a TJ max of 110c which is obviously stupid hot and sure enough is what's causing my fans to ramp to 100%.

Of course, this all meant it was time to take apart the GPU and re-thermal pad it as well as shunt-modding while I'm in there. Now I did the only silver paint method despite soldering my whole life as I was still too worried about the risk associated and the resistors are in a few tight spots I just deemed I couldn't do without being drenched in sweat after doing just one. I also mistakenly only got a single thermal grizzly thermal pad kit and not 3 like I'm supposed to (just reordered 3 packs and they come Thursday so I plan to redo this and likely add more silver paint if necessary).

So I only did the backplate and it was hardly enough but I made it work, just not to my expectation, but that's what I get for not reading I need 3. I taped around all my resistors, scraped the coating off of the ends of the shunts, and very carefully painted the actual resistor from end to end trying to make as much meaningful contact as possible. I also tried to make each one as consistent as the last obviously so we don't piss off Nvidia's power balancing.

Now, the results were interesting! I did this all at work and brought the card home and fired her up no issues. Under load, the memory junction temp was definitely better by a few degrees and didn't get as hot as fast, but it inevitably still gets to about 106 on a long gaming session or benchmark and that's what I get for not reading everything around buying 3 kits... Still an improvement and most importantly I can still use my card until then. The other thing is that is definitely helped despite me not being able to really gauge a power difference. Check this out:

Previous results with the 3090 maxed out in every way in my 9900KF build:

Port Royal: I scored 13 511 in Port Royal
Time Spy Extreme: I scored 9 334 in Time Spy Extreme

New results last night with my 5950x somewhat dialed in:

Port Royal: I scored 13 830 in Port Royal
Time Spy Extreme: I scored 10 775 in Time Spy Extreme

Obviously besides a higher score, but look at those sustained clocks!!! That's a significant improvement over my past average clock in both benches and just in general I could immediately tell it wasn't dipping down into the 1800MHz and 1900MHz range in some loads anymore which is great. This card literally would dip down as far as 1890 depending on the workload no matter what slider I maxed out. Now, the lowest I was seeing was 1980MHz and some points and that was super rare... also, lower peak clocks, but maybe that's a driver thing? Not sure, I was hitting 2200MHz before somehow granted it performed worse so maybe a moot point. I know HAGS for sure affects it, but I'm 90% sure I had it on in the first benches too.

What's weird to me is what GPU-Z is showing and I'm hoping I can squeeze just a hair more out of the card considering what it shows:

View attachment 2479088

This screenshot was taken after one of my earlier runs yesterday before I maxed the voltage slider (which seems to do literally nothing this gen and especially for this card) and also before Windows's HAGS was turned on (which effectively lowers my clocks a bit but gives me a pretty hefty performance gain), but my power slider was maxed in MSI afterburner. Notice that pin 1 and 2 are slightly different and during some benches, they can be usually as far apart as 35w as well as pin 2 is almost always .1v higher than the other... clearly I knew these would change so I thought they'd still stay the same, but again, I definitely saw a performance improvement related to power so it's far from a negative. From what I remember at the very least is that "Board power draw" shows no difference from stock as it should be around 400w to begin with, but I figured this would actually show less despite truly drawing more power.

Anyway that's mostly what I've got. I ran an insane amount of benches last night if you need any more to look at, but I'm really hoping one of the shunt mod gurus in here can give me an idea of what I'm dealing with, what my numbers mean, why the pins are different, how I can push it a bit more (by re-shunt-modding), and maybe give me an idea of what I'm actually pulling power-wise now. I hooked up the kill-a-watt last night and saw all of a 400w jump from when the benchmark would load (2% GPU usage) to when it started and went to 100% usage, so it's definitely there if not pulling more. I also now know with my current CPU settings and the card maxed, I should see around 670w from the wall give or take so I can use this for reference if I redo the shunt mod to find any differences.

Thanks for reading and thank in advance for your time and help!

Your 8 Pin#2 and PCIE Slot aren't painted well enough. Could possibly be a contact issue, scraping the edges/conformal coating issue or simply another paint layer needed.
8 pin#2 is easier to touch up (way easier) than 8 pin#1. With the PCIE slot pointing downwards, 8 pin_2 is the one on the far right of the power connector area, to the right of that square choke/inductor, and in an up-down orientation so easier to mod the edges (V shape PCB facing right, PCIE Slot pointing down, with GPU core side available), and 8 pin #1 is to the left of the inductor (in a left-right orientation, harder to mod because the right side is right next to the inductor). MVDDC shunt is by the V cutout below the 8 pins.

PCIE Slot has more components around it so take care and apply tape around that shunt. With the PCIE slot side pointing up, that shunt is at the top left. With the PCIE Slot side pointing down, that shunt is to the bottom right.

Also, soldering stacking shunts is rather easy. Since you're worried about lack of space, use 3M high temp Polyimide tape and just tape over the PCB. High quality 3M tape is much much better than cheap chinese clone or random kapton tape. 1/4" or 1/2" works fine (1/4" is probably better since you don't have to trim it as much to get it into narrow spots). That will protect the PCB very well. And you need a decent temp regulated soldering iron (65W or better, a TS100 is pretty baseline for this type of work), rosin flux (flux is essential), and good solder. You need to heat the edges of the shunt for several seconds before the solder will flow from the iron to the edges of the shunts. On the FE, since the shunt edges are 'lower' than the middle, you need to basically make two 'solder bridges' to fill that space, then flux on top of them, then put the new shunt on top, and then solder the edges together, one side at a time.

5 mOhm stacked shunts work well for this.

 

[TJD269]

Your 8 Pin#2 and PCIE Slot aren't painted well enough. Could possibly be a contact issue, scraping the edges/conformal coating issue or simply another paint layer needed.
8 pin#2 is easier to touch up (way easier) than 8 pin#1. With the PCIE slot pointing downwards, 8 pin_2 is the one on the far right of the power connector area, to the right of that square choke/inductor, and in an up-down orientation so easier to mod the edges (V shape PCB facing right, PCIE Slot pointing down, with GPU core side available), and 8 pin #1 is to the left of the inductor (in a left-right orientation, harder to mod because the right side is right next to the inductor). MVDDC shunt is by the V cutout below the 8 pins.

PCIE Slot has more components around it so take care and apply tape around that shunt. With the PCIE slot side pointing up, that shunt is at the top left. With the PCIE Slot side pointing down, that shunt is to the bottom right.

Also, soldering stacking shunts is rather easy. Since you're worried about lack of space, use 3M high temp Polyimide tape and just tape over the PCB. High quality 3M tape is much much better than cheap chinese clone or random kapton tape. 1/4" or 1/2" works fine (1/4" is probably better since you don't have to trim it as much to get it into narrow spots). That will protect the PCB very well. And you need a decent temp regulated soldering iron (65W or better, a TS100 is pretty baseline for this type of work), rosin flux (flux is essential), and good solder. You need to heat the edges of the shunt for several seconds before the solder will flow from the iron to the edges of the shunts. On the FE, since the shunt edges are 'lower' than the middle, you need to basically make two 'solder bridges' to fill that space, then flux on top of them, then put the new shunt on top, and then solder the edges together, one side at a time.

5 mOhm stacked shunts work well for this.

Hey thanks a ton for the reply. I've read this thread many times and have even read your posts on reddit regarding this so it's great to hear from someone who knows what they're talking about.

With that said, I obviously already painted so I think I'm already too far to go and solder on shunts and I've already done most of the work for the paint to function .

So I will paint another layer on 8 pin_2 and the PCIe slot and report back, hopefully allowing me to push the card a bit harder since she's still power limited, granted even at 550w it sounds like it still will be.

Thanks a ton for the reply and help!

 

[Falkentyne]

Hey thanks a ton for the reply. I've read this thread many times and have even read your posts on reddit regarding this so it's great to hear from someone who knows what they're talking about.

With that said, I obviously already painted so I think I'm already too far to go and solder on shunts and I've already done most of the work for the paint to function .

So I will paint another layer on 8 pin_2 and the PCIe slot and report back, hopefully allowing me to push the card a bit harder since she's still power limited, granted even at 550w it sounds like it still will be.

Thanks a ton for the reply and help!

Sounds good.
You can zoom in on the images to see the labels.

Even with shunts, you won't be able to exceed 550-600W, not because of TDP limits, but because of internal rail limits, possibly related to MSVDD / NVVDD rails that cannot be shunted. Kingpin owners can change the loadline settings or increase MSVDD voltage to help circumvent this when using the 1000W Bios. Strix cards that are shunted have higher internal limits for these rails already, so they can usually reach about 650-700W.

For your card, you want to not be limited by the shuntable rails or the main TDP limits (114% TDP reported to GPU-Z, and the rail limits of 237W GPU Chip Power, 162-175W for the 8 pins (the SRC limits actually control this), and 69W for PCIE Slot).

So ideally, you would reach a power limit at about "107% TDP Normalized", while the TDP limit is at 100% or below, with all the rails I listed above below the values I posted. That's when you know you truly maxed the card out and aren't limited by the main shunts, but by the firmware (Bios).

TDP Normalized isn't a total TDP like TDP% is. TDP Normalized reports on any single individual rail, vs its own internal power limit, as well as the MSVDD/NVVDD limits.

 

[lordzed83]

@Falkentyne I see ya know fair bit about mods so maybe You could tell me what resistors iw not painted good enough yesterday

Thats how it looks. As an actuall professional electronican i cant solder on FE board (id do on others) Cause my hads are shaky like an alcoholic in the morning so went with PAint even tho i have resistors in front of me but not expected to get Founders White crow Anyway
Thats how it came on front

Did not make descent one of back so

I did them 3 ones. Since its PITA to do with custom loop that was Side thing i tried when upgrading thermal pads and liquid metal. So any Ideas and Sugestions whats not connecting welcomed .

 

[TJD269]

@Falkentyne awesome, thank you sir, will report back.

@Falkentyne I see ya know fair bit about mods so maybe You could tell me what resistors iw not painted good enough yesterday
View attachment 2479185

Thats how it looks. As an actuall professional electronican i cant solder on FE board (id do on others) Cause my hads are shaky like an alcoholic in the morning so went with PAint even tho i have resistors in front of me but not expected to get Founders White crow Anyway
Thats how it came on front
View attachment 2479181

Did not make descent one of back so
View attachment 2479183

I did them 3 ones. Since its PITA to do with custom loop that was Side thing i tried when upgrading thermal pads and liquid metal. So any Ideas and Sugestions whats not connecting welcomed .

Just to clarify, did you not paint all of them or am I crazy in saying that it looks like your resistors are in different spots? Some of the electrical components, including the resistors seem to be in different spots after looking at your first picture.

 

[HiLuckyB]

I scrapped and painted all of my resistors last Saturday when installing a waterblock on my FE 3090. From what I can tell I gained about 30-50 extra watts, Which is good enough for my card to stay off the power limit 95% of the time. My card will only run around 2100-2115MHz in game's like Cyberpunk, Not the best card but it's not doing to bad. I forgot to take a picture of the other side before I put it all together

 

[Falkentyne]

@Falkentyne I see ya know fair bit about mods so maybe You could tell me what resistors iw not painted good enough yesterday
View attachment 2479185

Thats how it looks. As an actuall professional electronican i cant solder on FE board (id do on others) Cause my hads are shaky like an alcoholic in the morning so went with PAint even tho i have resistors in front of me but not expected to get Founders White crow Anyway
Thats how it came on front
View attachment 2479181

Did not make descent one of back so
View attachment 2479183

I did them 3 ones. Since its PITA to do with custom loop that was Side thing i tried when upgrading thermal pads and liquid metal. So any Ideas and Sugestions whats not connecting welcomed .

I don't understand what you're saying.
Did you do the ones on the back of the card or not? You circled them but I don't know what that means.
You must do all six shunts. If you skip any it throws the power delivery way off. None can be skipped.
Skipping "SRC" throws off everything. Skipping PCIE slot throws off MVDCC. Skipping GPU Chip Power can cause early power throttling and insufficient power draw....

Your mod is poorly done. You are throttling on 8 pin#1 early, and your PCIE Slot is also looking like it was not modded either.
Did you do 8 pin#1 ? The shunt in the first picture, on the left, is in the shadows and I can't even see it at all. This is a 3080, not a 3090 but I believe the one in the shadows is 8 pin#1. And the other two visible are 8 pin#2 and MVDDC (FBVDDC)

Either you didn't scrape the edges of the original shunt properly on 8 pin#1, or that paint you used is terrible.
That paint can be used for shunt stacking on flat shunts (1 watt 2512 shunts are completely flat and level), but is very bad for 2 watt shunts. 2 watt shunts have the edges below the middle, so the shunts you are stacking will not make direct contact. MG842AR silver paint will work better for this stacking temporarily, but longevity is not assured. This paint is carbon (graphite?) based paint, not silver based so its even worse for shunt stacking.

And you didn't stack shunts. You used this paint as its own shunt. This paint has too high resistance and is not suitable for painting as its own shunt. Only MG 842AR would work for this. This Black paint can only be used for 'stacking' 5 mOhm-10 mOhm shunts on top of original shunts, as an adhesive, and only if the original shunts are flat (1 watt) shunts.

FE and Gigabyte use 2 watt shunts (depressed silver edges, higher black middle).
Gigabyte, Asus and MSI use 1 watt shunts (flat edges, flush with middle).

And solder stacking shunts isn't a problem for your shaky hands. Just use kapton tape (3M) and cover up the board, then you are safe. Just don't use low quality tape. And remember to flux, and heat the shunt edge enough so the solder will flow. You need a solder bridge on the edges like in this video because the FE shunts are 2W shunts. (Note: the shunts in this video are 1 watt, evga, but he applied a solder bridge first, thats what must be done on FE cards).

For 1 watt shunts (much easier!!!) you can do the direct method---flux, then apply shunt, then solder one side at a time together. (16:00 in this video).
This method won't work on FE because of the lower edges, the shunts won't contact each other, must apply a solder bridge first.

 

[lordzed83]

I don't understand what you're saying.
Did you do the ones on the back of the card or not? You circled them but I don't know what that means.
You must do all six shunts. If you skip any it throws the power delivery way off. None can be skipped.
Skipping "SRC" throws off everything. Skipping PCIE slot throws off MVDCC. Skipping GPU Chip Power can cause early power throttling and insufficient power draw....

Your mod is poorly done. You are throttling on 8 pin#1 early, and your PCIE Slot is also looking like it was not modded either.
Did you do 8 pin#1 ? The shunt in the first picture, on the left, is in the shadows and I can't even see it at all. This is a 3080, not a 3090 but I believe the one in the shadows is 8 pin#1. And the other two visible are 8 pin#2 and MVDDC (FBVDDC)

Either you didn't scrape the edges of the original shunt properly on 8 pin#1, or that paint you used is terrible.
That paint can be used for shunt stacking on flat shunts (1 watt 2512 shunts are completely flat and level), but is very bad for 2 watt shunts. 2 watt shunts have the edges below the middle, so the shunts you are stacking will not make direct contact. MG842AR silver paint will work better for this stacking temporarily, but longevity is not assured. This paint is carbon (graphite?) based paint, not silver based so its even worse for shunt stacking.

And you didn't stack shunts. You used this paint as its own shunt. This paint has too high resistance and is not suitable for painting as its own shunt. Only MG 842AR would work for this. This Black paint can only be used for 'stacking' 5 mOhm-10 mOhm shunts on top of original shunts, as an adhesive, and only if the original shunts are flat (1 watt) shunts.

FE and Gigabyte use 2 watt shunts (depressed silver edges, higher black middle).
Gigabyte, Asus and MSI use 1 watt shunts (flat edges, flush with middle).

And solder stacking shunts isn't a problem for your shaky hands. Just use kapton tape (3M) and cover up the board, then you are safe. Just don't use low quality tape. And remember to flux, and heat the shunt edge enough so the solder will flow. You need a solder bridge on the edges like in this video because the FE shunts are 2W shunts. (Note: the shunts in this video are 1 watt, evga, but he applied a solder bridge first, thats what must be done on FE cards).

For 1 watt shunts (much easier!!!) you can do the direct method---flux, then apply shunt, then solder one side at a time together. (16:00 in this video).
This method won't work on FE because of the lower edges, the shunts won't contact each other, must apply a solder bridge first.

Gues best option for me is paying someone to replace resistors if i want to do it as i said I KILLED 2 cards with soldering due to shaky hands...
Maybe Ill buy another card and send this to someone to shut mod and sell spare one off hmm.
With my xxx hlth im off wortk 10 months so far... Im actually Trained electrician with yearts of expirience that cant do it anymore due to hand and spine problems.
Did all 6 but got like 40w gain guess the paiont is ****. And its 7 hours job to get gpu out with my external loop.

 

[mattxx88]

last night i shunted another 3090FE (thanks @lordzed83 )

a pretty nice board compared to my previous 3090FE, this new one con push 1300+ (the old ones only can +900) memory on air, this weekend ill dig its OC'in propensity deeper.

 

[lordzed83]

@mattxx88 do mine lol

 

[smud]

Do you guys put thermal pads on these resistors with using water block? I want to put them on when the whole set comes to me to test

 

[lordzed83]

Do you guys put thermal pads on these resistors with using water block? I want to put them on when the whole set comes to me to test

There is no space for pads I think

 

[Takukou]

ØC LAB的零售版本为R002

 

[Falkentyne]

ØC LAB的零售版本为R002
View attachment 2481436 View attachment 2481438

Anyone notice this?

 

[PopinFRESH007]

Hello All,
I've finally received my EK 3090 FE waterblock and I'm planning on shunt modding it while I have it dismantled to install the waterblock. I had a couple questions after reading through this thread over the past two days just to make sure I'm on the right path.

1) Anyone in this thread with a 3090 FE + EK Quantum Vector FE waterblock find any clearance issues with stacked shunts? I'm planning on stacking 10mOhm resistors on the 5mOhm existing resistors.

2) Looking at the part number that @bmgjet noted for 10mOhm, those appear to have been discontinued and replaced with ERJ-MS4HF10MU. This part appears to be pretty much the same spec but 2W instead of 1W. There was also a listing for part WSL2512R0100FEA18 which looks like the same specs but seems like it might be easier to solder? Any recommendations on either part?

3) If I'm stacking 10mOhm on all 6 shunts according to @bmgjet 's calculator the card should be pretty safely able to pull 525W total power when set at 100% power limit in Afterburner? 112.5W from the PCIe slot power and 412W from the 12pin. If I were to set the power limit to 110% in Afterburner, assuming other limits aren't hit and there is a load that demands it, the board would be able to potentially draw up to 577.5W, pulling 123.8W from the PCIe slot?

Thanks in advance for any advice. Also big thanks to @bmgjet that is a pretty useful calculator.

 

[Falkentyne]

Hello All,
I've finally received my EK 3090 FE waterblock and I'm planning on shunt modding it while I have it dismantled to install the waterblock. I had a couple questions after reading through this thread over the past two days just to make sure I'm on the right path.

1) Anyone in this thread with a 3090 FE + EK Quantum Vector FE waterblock find any clearance issues with stacked shunts? I'm planning on stacking 10mOhm resistors on the 5mOhm existing resistors.

2) Looking at the part number that @bmgjet noted for 10mOhm, those appear to have been discontinued and replaced with ERJ-MS4HF10MU. This part appears to be pretty much the same spec but 2W instead of 1W. There was also a listing for part WSL2512R0100FEA18 which looks like the same specs but seems like it might be easier to solder? Any recommendations on either part?

3) If I'm stacking 10mOhm on all 6 shunts according to @bmgjet 's calculator the card should be pretty safely able to pull 525W total power when set at 100% power limit in Afterburner? 112.5W from the PCIe slot power and 412W from the 12pin. If I were to set the power limit to 110% in Afterburner, assuming other limits aren't hit and there is a load that demands it, the board would be able to potentially draw up to 577.5W, pulling 123.8W from the PCIe slot?

Thanks in advance for any advice. Also big thanks to @bmgjet that is a pretty useful calculator.

You can use these for 10W:

https://www.mouser.com/ProductDetail/Panasonic/ERJ-M1WSF10MU?qs=%2Fha2pyFaduhOx12dbJfB0MsN%2FezM8qu6pargcXxcXkqNR6e9JV9E2g%3D%3D

but just buy the 5W shunts and stack them. It's better.

https://www.mouser.com/ProductDetail/Panasonic/ERJ-M1WSF5M0U?qs=js9DCdkuA2rWKXij%252BGW9Dg%3D%3D

you're going to be limited by internal rail limits (NVVDD, MSVDD, PLL) that you can't shunt or do anything about anyway, long before you reach the actual TDP limit. Only Quake 2 RTX will allow close to the real power limit, due to the age of its engine and it being purely path traced raytracing.

Make sure you use a power limit HIGHER than 100% if you want full power, otherwise due to weird stuff with "normalized" rails you can't shunt, a TDP% between 80%-100% won't give you any more power. Using 100% power limit will cause the "default" NVVDD / MSVDD limits to throttle you at 95% "TDP Normalized", which is a different thing than TDP%. But both can trigger a PWR flag. At 114% TDP, the "maximum" values for NVVDD / MSVDD / PLL will start throttling you at about 107% or so of Normalized. I don't think the "Normalized" rails can report a power value below the "default" limit for that exact rail (which is the 100% value).

Unfortunately, no one knows what these limits are exactly. I only what they do and how they work on this card, since I do own the card and put in my time testing.