[GribblyStick]

I am currently looking to upgrade to water cooling and I have a 3090Asus Strix OC.
Highest peak board draw I've seen so far was 516W and still being power limited.
I'm going to have to take it apart anyway for the water cooling, so I figured I might as well consider adding a shunt mod while I'm at it.
At those readings though, I have my doubts it would have much of an effect?
I could try with an EVGA BIOS but I can't say that sounds like a terribly good idea to me.
I want a useable card, I'm not shooting for benchmark records.

I'm planning on adding a back plate cooler too because that back plate is currently getting really really hot.
In Uniengine ultra I am getting up to 80 degrees on the core, so I imagine the memory on the backside is getting REAL toasty.
So that should give me a fair amount of headroom for overclocking.

 

[Cryptedvick]

Doing that will void the warranty of a very expensive card, for no more than a few percentage of extra performance, for a heck of a lot more power draw.
There are no advantages to shunt mod, unless you're doing benchmarks, trying extreme overclocking for that little bit of extra performance and Mhz. With shunt mod, people are hitting around 100, maybe 150Mhz more, at most, which translates to a very small, imperceptible difference in real world. Its already the strongest video card on the market.
If you want to overclock more efficiently, use the Curve in Afterburner and try to find the least amount of voltage needed for a set Mhz frequency. This will allow you to overclock the card and prevent it from hitting power limits (mostly).
On my 3080, I cannot hit 2000Mhz on stock voltage because I am seriously limited by power draw, but if I undervolt to 0.95v, I can hit 2010-2025Mhz stable, with it rarely hitting the power limit enough to lower my Frequency.

 

[mouacyk]

I would go for it -- enthusiasts are the nemesis of glass ceilings, and these PLs are very hard ceilings. I'm tempted to do likewise on my 3080, but I'd feel better knowing the following factors more:

1) How much wattage is the VRM really capable of handling, without degrading or failing
2) Durability of the "easy" shunt mod using conductive paint

 

[Xinoxide]

strix 3080 and 3090 are like 1000w / 1200w respectively. should be able to set upwards of 1.9 with direct control of the MP2888a ( if it is ) via i2c. also strix cards have unpinned through-holes for differential sense control via an analog EVC like the EVC2SX, ASUS ROG OC PANEL and just plain old fashioned hardmods.

Information like that doesn't float around these pages, too softcore here now.

edit: Personally, I do this whether I use the card regularly or not. I just don't run with tuned settings... for longevity.

Id like to be able to turn it back up later to play with it without worrying about it being in bad shape.

 

[SPL Tech]

Doing that will void the warranty of a very expensive card, for no more than a few percentage of extra performance, for a heck of a lot more power draw.
There are no advantages to shunt mod, unless you're doing benchmarks, trying extreme overclocking for that little bit of extra performance and Mhz.

I noticed smoother gameplay in Cyberpunk after I shunt moded my 3080. It helps some. He could use the silver pen and then his warranty will still be there.

 

[GribblyStick]

strix 3080 and 3090 are like 1000w / 1200w respectively. should be able to set upwards of 1.9 with direct control of the MP2888a ( if it is ) via i2c. also strix cards have unpinned through-holes for differential sense control via an analog EVC like the EVC2SX, ASUS ROG OC PANEL and just plain old fashioned hardmods.

Information like that doesn't float around these pages, too softcore here now.

edit: Personally, I do this whether I use the card regularly or not. I just don't run with tuned settings... for longevity.

Id like to be able to turn it back up later to play with it without worrying about it being in bad shape.

Well I don't know about 1200W lol. That is extreme overclocking territory and would require a chiller I imagine.
I don't want to go quite THAT crazy. I would be getting a 360/60 mm rad in push/pull and another 360/40mm either push or pull, we'll see how it fits lol.
my CPU is currently drawing up to 140ish (non overclocked, besides bas PBO), GPU I've seen up to 516, so I'm already looking some 700ish W to get rid of with roughly 1200W of cooling(with the custom loop). So I think I could only go up to 700 or so W for the GPU to leave some headroom for spikes and fans and whatever else might draw power. That kind of PSU wouldn't have a the best efficiency either I'm guessing and I'd rather have left over power than not enough.

In that sense I'd be be looking at what? 1.2 - 1,3 V at 60ish degrees under full load?
Currently I'm trying undervolting as suggested, but I'm getting pretty diminishing returns after 0,95V.
Sitting at 2025 GPU / 10501 (1312) Memory on 0,950V sitting around 76 degrees. I can get it to 2100 but it doesn't stay stable at that and keeps dipping, although it runs stable in that nothing is crashing. I'm guessing there is some thermal throttling going that I'm not seeing in GPU-Z?

so between watercooling and a shuntmod, Do you think I could hit 2200 / 11000? That might be worth it to me, anything less and I'm just going to run with the watercooling and call it a day.

 

[o1dschoo1]

Doing that will void the warranty of a very expensive card, for no more than a few percentage of extra performance, for a heck of a lot more power draw.
There are no advantages to shunt mod, unless you're doing benchmarks, trying extreme overclocking for that little bit of extra performance and Mhz. With shunt mod, people are hitting around 100, maybe 150Mhz more, at most, which translates to a very small, imperceptible difference in real world. Its already the strongest video card on the market.
If you want to overclock more efficiently, use the Curve in Afterburner and try to find the least amount of voltage needed for a set Mhz frequency. This will allow you to overclock the card and prevent it from hitting power limits (mostly).
On my 3080, I cannot hit 2000Mhz on stock voltage because I am seriously limited by power draw, but if I undervolt to 0.95v, I can hit 2010-2025Mhz stable, with it rarely hitting the power limit enough to lower my Frequency.

This is overclock.net not I'm scared to overclock.net.

Do it.

Edit. Put it under water first. See if it will do the clocks you want and if it hits power limit.

 

[Xinoxide]

The boost algorithm is dependent on temps first and foremost.

Undervolting will help it boost on its own and with help behind it.

locking her down to -177c load will keep the boost algorithm locked to the max freq but boost starts at like 30c or something low, cooler the faster it boosts.

If you can put under water I dont see why 1.3v and almost 600watt draw will be absolutelly fine, but will need shunts of course.


And the 1200w is just its highest-ish potential. 3080 just missing a handful of fets but still up there.

only other carrd that ate that much power cold so far was 980ti, and neither evga or asus made vrm like they did this time. :]

 

[cstkl1]

cooling first. wc etc
then consider the rest.

 

[SPL Tech]

a

 

[Cryptedvick]

This is overclock.net not I'm scared to overclock.net.

Do it.

Edit. Put it under water first. See if it will do the clocks you want and if it hits power limit.

Pall, last year I blew up my 8700k and Mobo while AVX stress testing, looking for a few more % of performance, and I've been overclocking for the better part of the last 15 years. Trust me, I know what this place is.

Since GribblyStick mentioned in the OP that he wants a usable card, not shooting for benchmark records, its not the wisest thing to void a warranty and potentially ruin a 1500 dollar card (or more) for that extra 4-8% of performance vs what he would have gotten from a fine tune Curve OC in Afterburner and maybe a BIOS flash.
There is literally no reason to shunt a 3090, unless you're planning on very serious cooling for it (since it will draw more than 600-700W) and shooting for some benchmark highs.
Remember, the card is already maxed out. You'll get maybe an extra 100-150Mhz from it, which translates to a few more FPS on top of the >100FPS you're getting in most games. Is this worth the extra 40% in power draw, higher temps, warranty void and potentially killing the card from a bad shunt mod? Common sense says no.
If GribblyStick doesn't really care about the card and what he spent on it, by all means, put it under a beefy water block, shut it and run 2250Mhz on it and enjoy a VERY fast card.

 

[Renegade5399]

Did it to my eVGA 2080Ti's after water cooling them. I don't think I will ever not do it again. It improved things greatly on the stock BIOS (1st BIOS was 280W, then update from evga set it to 338W). With the card thinking it was drawing less power (factor of 1.625), I was able to hold higher clocks. With the 338W BIOS I would get 2085-2055 with the typical game/rendering speed settling at 2055 once temps levelled. vRAM was set to 8000 (16000 effective) and forgotten about. Not really an issue if you got the cards with 16Gbit Samsung chips. I got mine through step-up and the serial numbers are sequential and both cards have this memory. I don't know if they failed 16Gbit (2080Ti spec'd for 14Gbit) testing or it's all they had. This was during the whole "OMG MICRON MEMORY IS SO TERRIBAD AND MY 2080Ti SUCKS" phase. Either way, I have never had an issue setting them to 8000 on an eVGA BIOS (XOC BIOS does wonky things with the timings and 7800 is the limit).

Average fps in CoD (MW/WZ or BO:CW) was about 150fps at 1440p high/ultra settings, RTX OFF and DLSS on Balanced. Perfect for my 144Hz display. I can't remember what Cyberpunk was as I have not played it since a few days after launch (I want the game complete and fixed before I play it through), but...nevermind, I found the "New Text Document.txt" on my desktop with the settings. 1440p, Ultra, RTX OFF, DLSS Balanced was seeing 95-125fps in outdoor/open areas and 100-145fps indoors. In both those games I run DLSS even though RTX is OFF because it speeds up the rendering of things or something and gameplay feels smoother. This is with the stock 338W BIOS from eVGA as, at the time, I stopped messing with other BIOS because none of them could support the memory properly. So: In my own personal experience there was a noticeable difference in performance vs no shunt while being able to keep the benefits of the OEM BIOS. In my case the original BIOS was a measly 280W because I went for the lowest class "A" GPU I could find within my means since I was going to water-cool them. Step-up with 2x1080Ti FTW 3's made these cards $680 each, shipped. You just have to be willing to void the warranty of a $1000+ dollar card (or in my case, 2 of them) and be able to accept the consequences of doing such a mod. Not that it needs to be said, but soldering on a couple resistors, beating the card until it dies, removing the mod and then attempting to issue a warranty claim is kind of scummy IMO.

Now last week I got "the bug" again and decided to check the 2080Ti thread here to see if there were any new BIOS files since the last time I visited. Well hello there GALAX HOF OC Lab BIOS that has all the features enabled (Voltage/Frequency Curve Editor & 2D clocks especially). Besides the profile loading issue with MSIAB and Xtreme Tuner (BSOD if you have it set to load a profile on windows load), it's been a great BIOS. An XOC style BIOS with all the limits removed, 2000W PL, but has the nice features? Yes please. Best part is it works with my vRAM and I can set to 8000. This BIOS makes the shunt mod unnecessary, but makes water cooling mandatory. I still haven't fired Cyberpunk back up, but I play CoD all the time. Now the shunt mod definitely helped, but this was the way to go. I did not gain very much speed. After some testing, the 2080Ti I have in now just tops out at 2085, no matter the volts it just does these weird bursts of light flashy things in Heaven past that and other tests fail or are wonky. GPU memory speeds set to stock for stability testing BTW. So with the new BIOS I set it to 2085/8000 @ 1.125v. 29°C idle and a max of 52°C extended load. Once at that 50°C mark, it clocks down a bin to 2070. With Xtreme Tuner if you don't check the box next to the GPU VOLT slider, normal features like the binning down for temps and 2D down-clocking still work. I set the voltage using the GPU Volt (Offset %) slider. FPS in CoD is now in the 150-200 range and I shaved a ms off the "GPU Time" counter in game (some Reflex thing). That many fps gained for only 15MHz clocks? Not a chance. So, I flashed the card back to the OEM BIOS and ran some tests with GPU-Z running. First Heaven (1440p all maxed), then Superposition (8k) and there it was. PerfCap Reason: Pwr. Checked the Max Board Power draw and saw 373W. So I was smacking the 380W power limit even with the shunts. Keep in mind that 373W read by the card is actually 606W. You have to take the read value and multiply it by 1.625 (0.008ohm on top of the 0.005ohm). Flashed the GALAX BIOS back on and ran some tests. No Pwr PerCap Reason this time. Now, this is where you have to sit and think about what it is you want. The shunts allow you to use stock BIOS with there still being a chance you hit the power limit while voiding warranty. The XOC style BIOS simply allows for unleashed performance without needing to do mods. There has to be a BIOS like that publicly available for your card however. Those can be slow coming. Since you have an ASUS Strix OC 3090, your chances are a bit better that an XOC BIOS will become available to you. Additionally, you are going to need a serious supply of power. I run my card on it's own dedicated 750W supply. Some leftover mining HP server supply and breakout board I had. With all the limiters removed I saw a Max of 444.8W in GPU-Z. Doing the math on that comes out to 722.8W. Did it actually pull that for a split second? Maybe, but the point is the power draw goes up exponentially.

So, it comes down to personal choice:

Hard mod a card and lose warranty, but gain a good bit of performance and keep the OEM BIOS.

-Or-

Soft mod a card, but lose all power efficiency and possibly other features since XOC BIOS are almost never as nice as the GALAX one.

 

[GribblyStick]

The 1200 W was in terms of water cooling potential. As in, an estimated 200W per fan with two 360 rads. One 40mm the other 60 mm. Or maybe two PE because can only do XE360 in push as opposed to push/pull (unless I sidemount it). So all in all about 600 per rad for 1200. Looking at an EK XE360 and an EK PE360 at the moment, aquacomputer rads look really nice and just at a glance way higher quality, but from the reviews I can find they seem to perform pretty badly by comparison, though given the evident quality it doesn't make any sense to me.

I was thinking of daisy-chaining my 5900x, the 3090 strix oc (with potential shuntmod), backplate cooler and some 3800 or 4000Hz RAM sticks (4*8 probably), whichever kit with the tightest timings I can get my hand on. I was expecting maybe 700for the GPU if modded with another 150ish let's say for the CPU. Then with the odd spike we're looking at maybe 1000W tops for an instant.
I was seeing 516 already for JUST the GPU, it was just momentarily but hey. According to GPZ-Z anyway that was at 138 %TDP somehow.

Maybe I'm wrong, but I feel more comfortable with the idea of adding a few shunts than flashing the bios. I'm not too concerned about loosing warranty, doesn't mean I'm overly keen on frying my card either though I play mostly single player games at max settings these days, meaning that even at 1440p with the strix OC I am barely holding 60 frames in some games. (Odyyssey G9 so more like "1400p"). Getting a few more % is the difference between running a stable 60 fps or not. I'm not going to be anywhere remotely close to 240 frames unless it's an older game, but that's fine, I've been gaming at 60 Hz for so long I really don't need more. RTX and friends matter more to me and that requires every last bit of juice I can (reasonably) squeeze out of it.

What I'm getting here is that I'll get a loop running and see what that does first before messing with the card. If 2100 is the about the wall even with the average shunt mod, it might not be worth it.
Thanks all, now to go down the watercooling rabbit hole.

 

[Xinoxide]

There are multiple ways to attack it. Not everyone wants to inch up to 1.2v, Like myself. Very few are going to start at 1.55v and finding cold bug first thing either.

The potential is there for you to find out, its in the extra cost you paid.

 

[WMDVenum]

Deleted. Didn't read before replying.

 

[kiriakos]

I noticed smoother gameplay in Cyberpunk after I shunt moded my 3080. It helps some.

Yes 3080 this is a slow as hell VGA ... two resistors trickery helped you? and you are saying this with out offering of proofs ?
I do notice smooth game play only when servers are not full of LAG due large amounts of players.
Everything else, it is in your imagination.

 

[Renegade5399]

I noticed smoother gameplay in Cyberpunk after I shunt moded my 3080. It helps some. He could use the silver pen and then his warranty will still be there.

The microstutter issue with CP2077 was real. My card did it too. I fixed it by using the curve editor in MSIAB and setting the same frequency, but at a higher voltage point. This made it so the card was not binning down due to current draw. You adding the shunt mod had a similar effect, thus the smoother gameplay. The engine in that game (PC version) had a real issue with Boost. It has supposedly been patched out, but like I said previously, I'm not playing it until it's done (which may be never, I know guys, I know ).

Now a point of ethical disagreement (friendly discussion of course, not attacking you): Using the pen or soldering on resistors is the same thing. It's still a physical mod. Wiping off the conductive paint after killing the card and then asking for warranty replacement is dirty. It's that kind of that makes it so the rest of us cannot have nice things. I had a pair of 7900GTs back in the day that I pen (pencil actually) modded. Whist benching Aquamark (still have a top spot for 7900GTs in WinXP on HWBOT), both cards blew a canister cap off their boards. Despite being young, dumb and broke, I did not warranty them. Since they were not SMDs like these new cards are, I ordered the beefiest replacements I could find that would still physically fit and repaired them myself. Please don't cheat the OEM if you can avoid it. It's publicly available knowledge their profit margins shrink more and more each generation with the 3000 series being the worst yet.

Yes 3080 this is a slow as hell VGA ... two resistors trickery helped you? and you are saying this with out offering of proofs ?
I do notice smooth game play only when servers are not full of LAG due large amounts of players.
Everything else, it is in your imagination.

How can you have been here for 10 years and still spout off some dumba$$ like that?

...two resistors trickery helped you?

That's what he said. Reading hard for you?

and you are saying this with out offering of proofs ?

Without expensive video equipment to record modern monitors or the video stream, one cannot simply "offer proofs" of microstuttering being fixed.

I do notice smooth game play only when servers are not full of LAG due large amounts of players.

That's cool. CP2077 is a single player game there champ.

Everything else, it is in your imagination.

Like having an above average IQ is in yours bud?

⌚ 2️⃣🛑 your 🐂

 

[8051]

Pall, last year I blew up my 8700k and Mobo while AVX stress testing, looking for a few more % of performance, and I've been overclocking for the better part of the last 15 years. Trust me, I know what this place is.

Sorry to hear about your motherboard and 8700k. Were you doing the AVX stress testing with Prime95? I hope not since that's what I use to test my 9700k!

 

[8051]

Additionally, you are going to need a serious supply of power. I run my card on it's own dedicated 750W supply. Some leftover mining HP server supply and breakout board I had. With all the limiters removed I saw a Max of 444.8W in GPU-Z. Doing the math on that comes out to 722.8W. Did it actually pull that for a split second? Maybe, but the point is the power draw goes up exponentially.

What kind of power draw does your modded 2080TI usually see while gaming? So you have two PSU's for your system? One for the motherboard/CPU/SSD/HDD and another dedicated to the video card? Does your case have the capability of mounting two PSU's? I don't know if I'm out of touch but even 444Watts of power draw for a video card sounds like a lot, much less over 700Watts.

 

[SPL Tech]

Now a point of ethical disagreement (friendly discussion of course, not attacking you): Using the pen or soldering on resistors is the same thing. It's still a physical mod. Wiping off the conductive paint after killing the card and then asking for warranty replacement is dirty. It's that kind of that makes it so the rest of us cannot have nice things.

It depends on what kills the card. If I power mod the card and then burn up the power supply ICs and coils, yea that's on me. But if I mod the card and one of the VRAM chips takes a dirt nap, there is no reason why I shouldent be able to warranty the card as my mod had nothing to do with the fault. And before you stick up for the manufacture too much, remember they are doing the exact same thing in the other direction. Pretty much every manufacturer except Evga will void your warranty simply for doing a repaste or adding a water block, which is complete BS. Intel and AMD dont void your warranty for adding aftermarket heatsinks with whatever paste you want. You could use toothpaste and they would still warranty the card.

Also, unlike moding my card, striking a warranty for adding a water block is flat-out illegal. It is actually a federal offense.

Magnuson–Moss Warranty Act - Wikipedia

en.wikipedia.org