How insanely tone deaf is it for Nvidia to cheap out on something that can only cost them a few cents to do right? That's just Nvidia making it crystal clear what they really think of their customers.
Even in mass production the per-unit cost of a good thermal pad like Gelid or Fujipoly runs $20-40 for a sheet the size of trading card. Like the stuff that has similar heat transfer as thermal paste. Seriously, look for thermal pads with high conductivity on Digikey or Mouser. Using more expensive stuff would just increase the overall cost for the GPUs for not much benefit to Nvidia if it runs within spec. Yes there is a significant difference with using better pads. We’re talking about a company whose revenue is in the tens of billions every year. They’re all about the money.
@@rotor13 rather similar to how i make my sandwiches, i focus on the bread. just two slices put together and its ready to eat, no other ingrediants, sometimes i add a slither of butter as a treat. Nvidia have focused on the pcb and the cooler and neglected the ingrediant that makes them work together when charging consumers enough money they could buy a car for. regardless of how much it would cost for efficient thermal pads, it comes across half-arsed for a crucial part imo. those temps are a joke. but i suppose less life expectancy is a good thing in todays world, if you'll need to buy another card sooner. products are no longer built to last. they just managed to meet their 'in spec' requirement, despite it being a poor result and costing a few grand to consumers. i would like a little more consideration for consistant quality throughout the product when i drop that type of money.
they were forced to use GDDR6X memory at fast clocks just to beat AMD's Infinity Cache with GDDR6 ram. Because of that, it makes the gpu(in particular the vram) run hotter just so they can get the halo card into the market to get mindshare in purchase. Cheapening out on the thermal pads is just whatever they had in stock. It's still the cheapest RTX card.
For the cost of the GPU, the paste and pads 'SHOULD' be set correctly. The fact this wasn't, doesn't worry me so much, but makes me think the manufacturer no longer cares about the consumer.
Stuff is not made to last, it's made to cover the warranty period. After than, longetivity is actually a con for the manufacturer because selling more stuff gets harder.
@@MikkoRantalainen We are stripping the planet of valuable resources just to scam ourselves. Meanwhile the government keeps on talking about carbon bs that has little to do with the destruction of our planet or resources. We need to completely redesign our constitution as many new exploitative practices exist today that our forefathers never saw coming like medical tyranny, paid protestors, corrupted lobbying and severely corrupted political parasites.
@@ricky4673 For Nvidia it's pure and simple business and mass market bulk production (AMD as well for that matter). This means they will as often as possible try to get away with the least expenses as long as the main criteria for their core consumers are met. And they are. Them saving on pads when producing millions of cards, equals to big savings in the grand scheme. What this also does, is ensure that their business partners like EVGA, ASUS, etc can make very minor and easy upgrades and still see significant improvements, which enables them to market their products more easily without having to spend excess resources on r&d for every new generation hardware. Changing pads and addind a few better fans can easily drop temps by 15-25 degrees on average, which means they can more easily increase the power limit for overclocking, and overclocking means they can see more benefit from using higher grade memory components etc. Basically Nvidia and AMD tries to make the upgrade path as simple and efficient as possible for their business partners. After all, these are the partners that buy the vast majority of the overall graphics cards (outside of stock pc retailers). Making their partners' lives easier is essential for Nvidia. While this business approach has similar symptoms to something like planned obsolesence, it doesn't mean the outcome is the same. Most people who expect great performance and plan on maxing out their pc on a regular basis, are more likely to go for a 3rd party graphics card in the first place, like a Strix card which had the pads used in this vid.
This would have been a lot more useful if you actually gave a link or something to the pads that work well, rather than just saying "the grey squishy ones"
Once the plate is off and the chips/GPU are all cleaned, put a small amount of modelling clay on the chips and then put the backplate on. This will compress the clay to the exact thickness that exists between the components and plate. Carefully remove the plate and the compressed clay can be accurately measured. Pick the pad thickness based on that measurement.
@@RedHaloManiac95 yes, the modeling clay method is ok but it’s better to be on the thicker side rather than just perfect to provide pressure to provide maximum contact and thermal transfer and being under pressure the pads will eventually thin out slightly
This mod is mandatory IMO. When I opened my 3090 I have never seen such a poor job done at cooling from factory. Absolutely shocking. If you read around on this subject you'll find a guy who did some investigating and found that this memory starts to degrade at a much lower temperature than nvidia would have you believe. Pretty much every 3080/90 founders out there is slowly dying if you don't do this mod.
Out of curiosity, does AMD have this problem with their Video Cards? I want to know because it would be good for influencing what I should buy in the future.
The cynic in me says they did this so miners wouldn't be able to sell second hand cards and scare people who buy used from miners when these cards start failing
gelid EXTREME 2mm pads work great for this, gelid 2mm ULTIMATE are too big, i think its because the extreme are squishy, i dropped around 20-25c on my 3080 FE from doing this mod
@@rickynetherton2823 if you're careful yes, I used a 2mm and 3mm of those sizes and padded the front and back of it. I have a photo of my pads applied I could send if that would help guide!
Same temps with Frosty thermal pd 12.8 posted earlier above but saves your wallet. For FE i bought 1.5mm and dropped my temps identical to those pads. Works even better than Gelid Ultimate which I have no idea why. I heard Gelid Ultimate is too thick or doesn’t compress.
This is all well and good, but there's going to be a sizeable amount of people who just aren't technical enough to replace the pads. These GPUs should never have been released like this.
@@bepbep7418 not in a fairly regular lifecycle of 2-3 years. But that's not a good point to make. The cards running as hot as they do is a design flaw manufacturers can fix with literal cents.
@@noneofyourbusiness4294 but will customers buy new gen cards next year if their old cards work fine? especially today when you only need an upgrade if you're going from HD to 4K...Also it's not cents, buying a set of good pads could take $100 and more.
Exactly, it is totally unacceptable to sell cards like that, especially for what they cost. Modding should only be for those who want to exceed oem limits, it should never be necessary to do so, just to use the card under designed specs.
@@lukaogadze5071 Off the top of my head no, but maybe one of the Titan series? But honestly I don't think any have been quite that much before. At least not MSRP.
Jay launchin a GPU benchmark : "As anyone know, the RTX 3000 series 3080s run very hot" Jay after 10 min of GPU benchmark decides to touch the GPU : "Oh it's hot !" This man will never change ^^
I done the same when I first got a 980 Ti Classified a long while back. 80c on gpu, hotspot 92, vram/vrm around 100. I still touched the backplate though... and lemme tell ya, ya DONT wanna touch that thing when its running full bore.
@@ericspda I for one have no clue how come they can survive so easily in 78C in the warehouse :D Tbh I need to always convert with google from retard units
@@krzysiekj2522 I get it’s the cool thing to make fun of something you don’t understand, I just don’t see why people get so angry about a slightly different unit. It shows the same information with identical accuracy and typically higher resolution. Is it really that insane to consider a reasonable outdoor temperature as 75 degrees instead of 24? It’s all arbitrary anyways, it’s not like you frequently experience your body literally freezing or boiling, so it’s a bit confusing why water at standard atmospheric pressure is considered a gold reference everything should base off. Maybe you can educate me.
For anyone who wants to do this, make sure you actually measure the thermal pads using calipers to ensure you have the right size. some cards have 1 size pads, some have 2 different size pads, and some have 3 different size pads. you definitely want to make sure you have the correct size, saves you the trouble of having to re-open your card couple times and trying to figure out which pads might have been to big. just measure the the original pads, its better than guess work
my 3080 has 5 different thicknesses 0.5, 1,1.5, 2 and 3mm basically every single thickness you can get is on my card, thank god I found someone with the exact same card that did all the hardwork of measuring for me.
Best results i know - one russian miner did change RTX3080 thermal pads to 1 mm copper plates and memory temperature go down from 100C to ~66C. Not sure if it's his own idea but he did tried diffirent types of thermal pads/grease and all of them is nothing compare to copper.
@@NathanMc since when pretty sure as soon as you remove the pads unless you put everything back as it was you may not get your RMA accepted, depends on the party doing the RMA ofc but still that has been my experience in the past.
@@cr4zyw3ld3r In the us if they dont accept your RMA you just take them to small claims. They dont even show up to small claims since they know the laws not on their side so you win by default. Got MSI to refund me on two laptops that way.
*WARRANTY VOID WARNING!, NVIDIA STATED REPLACING THE THERMAL PADS WILL RESULT IN ANY RMA/RETURN BEING DECLINED.* Tens of people have already been told that doing this voids you warranty, so do it at your own risk. NVIDIA used crappy greasy cheap thermal pads and sadly if you replace them you see literally 15-20 degree decreases even. But you will also void your warranty. Just letting you all know!. Be safe!.
Ok, so I ran Time spy and fur mark, but my memory junction only reached a max of 96C, of course I am running stock out of the box EVGA FTW3 3080 TI. So it would seem as long as I leave at stock, I'm probably fine?
@@bowdiddly6172 yeah you are probably fine but you might be able to lower temps a little bit swapping pads and paste if you want. This problem is mostly a big deal for the Founders Edition cards. I’m not really aware of widespread issues on any of the AIB cards
If you’re on a budget and use Amazon there is a strong product to check: Frosty thermal pad 12.8. If you are on a super budget i recently bought some 6 W/mk pads in amazon from Outus store or something . Its like $10 for 30 67x20 pieces. I am not sure about the Outis ones just yet but they claim to be 6 W/mk which is still way better than stock pads. Id use those for the backs and use Frosty for the front as they are 12.8 W/mk
I had the same issue with the pads and solved it with a pasta roller. Rolled my 2.0mm pads for a bit and stuck them where i needed to use 1.5mm. Worked like a charm
I used a piece of glass (technically a small mirror) to do the same, simply squished them down Hard to get correct height, but sometimes its all you have left
The Thermalright Odyssey Extreme worked wonders for my 3080. Saw similar drops in the 20C range. Amazing that manufacturers skimp on $10 in better thermal pads for cards that are $800+.
1:55 they are also "flip chip ram" so most of the heat is dumped away from the heat sink anyway... so thermal pads on the "tops" actually makes more of a difference than most think it will.
*WARRANTY VOID WARNING!, NVIDIA STATED REPLACING THE THERMAL PADS WILL RESULT IN ANY RMA/RETURN BEING DECLINED.* Tens of people have already been told that doing this voids you warranty, so do it at your own risk. NVIDIA used crappy greasy cheap thermal pads and sadly if you replace them you see literally 15-20 degree decreases even. But you will also void your warranty. Just letting you all know!. Be safe!.
@@yakivpopavich That depends on where you live, they can't void the warranty for opening up the card in the USA. That shit sucks balls for some countries to let them void the warranty just for opening it up and making it perform better and cooler.
@@Apollo-Computers it makes sense because if the thermal pads potentially had something to do with the failure of the card Nvidia can't be expected to replace it because it isn't their fault that you modded it
I found that you need to first apply the pads to the card, NOT the cooler. The contact of the pads made best on the memory transfers to the cooler better. Not sure why...but lots of trial and error confirmed that every time I placed the pads on the memory first, it did better.
Thanks! This was really key. I was getting good results, but not what I expected, especially on Core temps. I just did this, with the same Gelid Extreme 2.0mm I used (opened and squished like 20 times in a row :( ) and it was GREAT. With undervolt to 1980@887 and +1000 on memories I was able to get almost 21k on air with the stock cooler. Paired with a 5900X on a skechy heatsink because I'm cleaning the liquid loop.
Rougher surface. Plus, one side is usually stickier (on purpose) than the other. And you want that side to be on the chips. When you apply them with your finger, you can get more air bubbles out. Then when the heat sink goes on, it's usually flat enough that it's not an issue.
You can simply measure the pads and if they are 0,8 you’ll need 1mm, if they’re 1,25 or so you’ll need 1,5mm because those pads where obviously squished a bit. Also not too hard, not too much silicone, not too soft. As for brands, Fujipoly 11 seem to be very good.
First thing I noticed with my Ventus OC 3080 was the insanely high VRAM temps. Bought some thermal pads and replaced only the pads around the die on the memory modules and mem junc. temp literally dropped by ~30C (from 108-112C to 80-82C)
PSA for the 3080 FE. The pads to use are the 2.0mm Gelid Extreme pads on the die side. On the back side you can use 3.0mm of Gelid or Thermalright. The reason why you have to use only Gelid Extreme is to Jay’s point, you really need 1.65mm but the Extremes are specifically soft so they can squish down to that size.
Yeah I used 2mm Thermalrights and I had to open it back up to squish them down more. I also applied extra paste to the gpu to help contact there. Gelid Extreme were out of stock at the time but they're definitely a key part of the mod.
@@ShimmeringSword yup I did that too got 3mm and 2mm and had to squish them down to get it to touch the plate. For some reason the noctua paste just couldn’t cool it good, so I got the kryonaut one and it helped. Temps don’t go past 82 on the vram. On a really hot day it got to 86, but that was at the peak of summer and not a well ventilated room. Definitely worth it. Got the gelid ones because I thought the thermalright ones were not going to work, but I ended up just keeping them and storing them away in case I gotta replace them down the road
My Vega 56 is the same way lol I was so shocked when I unboxed it because the backplate only covers the PCB. The heatsink is almost literally twice as long!
@@t1995roger10 I know somebody made a little 3060, but temps should still be fine at 3080 power levels. My rule of thumb is 150W per slot at full length, as the 1070 katana is just a tad warm.
From what I have seen from other videos and browsing through reddit, it seems nvidia set the fan curve to memory temps rather than gpu die temps so if you do this mod and see gpu core temps rise then it is totally normal because the fans aren't spinning as fast due to cooler memory. So just create a new fan curve after you do the mod if you are worried about the gpu die temps.
I did this on my 3090 a few weeks ago with good results, vram temps dropped by a lot but my core temps were about the same or a little higher by maybe ~2C. I think that’s because the core can push harder now that the memory doesn’t throttle so much. My VRAM temps dropped by around 20C, which was a huge relief! The old paste they used on the GPU looked like it was 10 years old and the thermal pads were pretty bad so I swapped the pads for Thermalright Odyssey 1.5mm and I repasted the die with Kryonaut (and used way too much I’m sure). As always YMMV but it’s worth it given the awful stock pads and paste
"It's within spec." Well, so is an Indy car that runs an incredible race and wins. It ran completely within spec...on the edge of its limits and is a million dollar paper weight in just 500 short miles. But, it ran perfectly within spec.
I like to run my ranger, C230, and Accord engines on the lower end of the spec. I be sitting there on the hottest day in summer with the A/C pulling the cabin temps to ice box.
The contently replace parts in the indie car. The start of the season car and the ending system could have had multiple components replaced to keep it in top racing condition
OK Jay ... so on the 3090 you used 1.5mm and that worked. However, you then said you used "really thick pads on the backplate". How thick does "really thick" mean? Is it 1.5mm, 2mm, 3mm? What's "really thick"? (I don't use Facebook or Twitter so could you kindly please reply here). I need to know because I am ordering what I need (normally don't do DIY on computers and have no pads), since I am seeing the 100C temps on my 3090 and need to fix this.
I did this to both my 3090FE cards back in April/May and saw a massive reduction in VRAM temps. I used 1.5mm Thermalright 12.8w/mk pads. I even tried the stupid expensive Fujipoly 17w/mk pads, but the Thermalright pads were actually slightly better. I read the the fibre/fabric pads than Nvidia uses are only rated for ~ 3 w/mk. That's shocking. £1400 each for the 3090 FE cards and they skimp on the thermal pads?
I drilled into the backplate of my 3090, drilled cavities into the backplate in between holes and ran 1mm heat pipes directly over ram module on the back plate, applied better thermal pads. Then i created a heat pipe array on top of the back plate and soldered a second copper backplate on top of that, then used thermal adhesive to glue on a bunch of aluminum heat sinks. Full memory overclock, +200mhz to GPU, 78C on ram after 1 hour of rendering with full ram usage!
So they're charging all-time high prices for gpu's and they can't even put decent performing thermal pads and paste in them? Seems like they could spend the extra dollar to put nice pads in them
That's the ironic thing though, this time around the paste was really good. That made it even more infuriating to have to take my 3090 FE apart to replace the pads which got me a near 30 degree drop for the memory junction temperature. I used Arctic MX-5 paste for the core and it ended up around the same decent temperature of the stock paste, about 45C idle with "prefer maximum performance" on and only as high as 68C rendering and 71C gaming.
@@Mangokr Make sure the measurements are EXACT. Odyssey's dont squish as much, so people have seen hotter temps as well because they arent high enough in certain spots, or too high. Thats generally why people suggest the EXTREME as the squish factor is better. a 2mm squish will fit where a 1.6-1.7 is needed, where as 2mm Odyssey wont, and lead to higher temps/bad contact. Etc. Edited. Not ultimate. Gelid Extreme. That is the one with better squish factor.
What I did was add an "IHS" onto each memory package. The actual dies are absolutly tiny, about 3x3mm and the theory of installing an IHS is it spreads the heat out allowing more thermal pad to do its thing. If you go onto ebay, you can type in copper cpu shim and that will return thousands of results. With a little bit of cutting and filing the shims fit perfectly on top of a memory package. I never changed the stock pads on my GPU, just added shims and the temps dropped from 106c mining to 85c mining.
for those of you doing this mod......... try putting thermal PASTE on the memory modules and then putting the new higher quality thermal pads on. It made a huge difference on my card for the memory temps.
Warning reader: I used 2mm and 3mm gelid ultimate on my card and my temps were worse. I tried a bunch of different variations too. The only thing that works and you dont have to mess with it over and over to get everything right; is using gelid extreme 1.5mm on the dye side. then u can fiddle with 2mm or 3mm on the back to find out what works best, i think the back its meant for 2mm though. its also extremely important that you get the gelid extreme pads exclusively, and not any other brand or style on the GPU side. gelid ultimates did not work better for me, in fact they made things worse.
If proper thermal pads drop the RAM temps by 20c, you can expect that to mean that the cooler is now handling more cooling load than it was originally intended for, so it is likely that the other components that need to be cooled will be cooled less efficiently. Probably a good idea to add an extra cooling fan blowing straight on to the card in that case (but be sure it's in the correct orientation so that it doesn't fight against the GPU fans!)
I would have liked to know, if the hot VRAM is only on the backside or ALSO on the front. In other words: Can i improve VRAM Temps, just by changing the ones on the Backplate? My reason for that assumption is the 3080 has only 1 VRAM Hotspot which can be cooled better by placing some Pads on the backplate.
i just checked on my gigabyte gaming oc 3080, and its just fine. up to 84° on memory with 66% fan, at 100% down to 76°. also core and hotspot dropped 10° when going fron 66% to 100% fan.
@@od_static what's the ambient temperature in your room like? In an air conditioned space of about 72° f my gigabyte gaming OC was hitting about 90 95C on the memory modules with the fan at about 80 90%. Yes it's technically within spec but the cooler your card stays the longer it'll last. Gigabyte is pretty known for using cheap leaky thermal pads. Yes I said leaky lol somehow their thermal pads leak oil after a while. Now under load I'm hitting around 75 85C on my memory and that's worst case scenario. It was also an excuse to put better paste on the core and that improved my temperature is another couple degrees too.
*WARRANTY VOID WARNING!, NVIDIA STATED REPLACING THE THERMAL PADS WILL RESULT IN ANY RMA/RETURN BEING DECLINED.* Tens of people have already been told that doing this voids you warranty, so do it at your own risk. NVIDIA used crappy greasy cheap thermal pads and sadly if you replace them you see literally 15-20 degree decreases even. But you will also void your warranty. Just letting you all know!. Be safe!.
Thank you so much for doing this video. I checked the temperatures on my Asus 3080 strix OC and is water cooled and was getting around 95C. Changed the thermal pads to thermal grizzly minus 8 pads and now the temps have dropped to around 65C. Huge improvement thank you
I wish Asus would once again have quality products... Any company that you can do a basic mod and see MASSIVE improvements is a company I would always second guess before considering purchasing any of their products :/
@@ihatesignupsgrrrrrrr I'm very happy to buy Asus products in the future as asus are not the only company this mod works really well on. At the end of the day the temperatures were within the spec of the memory.
Use my 3090 for mining and saw a 25C drop by upgrading the pads and adding some copper NVME heat sinks over the backplate. Have an old Gentle Typhoon A14 (?) blowing over them. Works like a champ.
Front and back side Gelid mods on the FE is a revelation. The stock pads are made of crumbly fail and frankly the stock die thermal paste application wasn't great either.
I watercooled my Gigabyte RTX 3080 with the Bykski waterblock with active backplate a few months ago. I used the blue thermal pads that came with the block. During mining I went from 110c(throttling/perf capping) with memory OC, to 90c. THEN I bought Gelid pads and just installed them this weekend, and I dropped down another 30c, to 60c with +1350 on memory. It's insane how much the thermal pads matter. I used 1,5 mm both on the front and back, for those wondering.
Ek pads are pretty bad due to their thermal conductivity in my experience. I recommend thermal pads by gelid. Also you could use thermal paste to fill in any gaps if you think the 1mm is too small and don't have any 1.5mm, or if the gap isn't exactly 1.5mm and is actually a bit larger.
I literally just did this to my 3080 Ventus and got exactly 20*C cooler mem temps and with only artic silver 5 paste got 8*C cooler core and I must have done a killer job with my paste because my hotspot is right on my core temp which was a 14*C drop! Also to note my ventus 3080 requires 2mm pads from the factory and I also had enough pad to replace the vrm and capacitors pads as well!
Since i switched my 3090 Eagle to the Bykski block, i have changed the pads by default and they seem to be decent. Will test a bit more and see what i can get, i still have the hotspot numbers from the aircooler/stock pads.
Problem is there's a lot of conflicting information out there about whether this voids the warranty. During these times where it's one per customer and I wouldn't be able to replace it myself, I'd value the warranty over the improved temps. You never know, your GPU could die of another cause beyond the crappy thermal pads and then you'd be out of luck...
Easy explanation of "but it's in spec": so is running a car engine at redline: that's the limit, you're running it at the limit. So... wanna try running your car to redline consistently over the life of your vehicle? Yeah, that's not a good idea. You can run it to redline, or even beyond under load, but it does cut into the life of your device.
I understand the underlying logic your analogy, however, it is flawed. You are comparing a mechanical engine with hundreds of points of mechanical failure, each of those parts with hundreds of points of failure and under multiple different amounts of forces and stresses. GPU mining is something everyone hates, but, as multiple outlets have covered and shown, has no tangible effect on the card. GamersNexus showed that a card mining at 100% utilization for ~4 years straight versus one that was basically new in box, were performing identically to each other. Will the GPU one wear out before the other? Seems logical to assume, but there is zero evidence to suggest it to be so. Electronics are robust and the stresses they experience aren't the same as mechanical stresses. Running a normal consumer car at redline for say 10 minutes straight will blow the engine, if I run my GPU at 100% utilization for 10 minutes and it fails, it isn't my fault.
@@johnlahey2157 100% utilization and reaching the max thermal capacity limit are not the same thing. if i run my sound equipment over 75% it's not sustainable and will overheat. similar to "boost" clock on cpu's. there's a reason why they don't just clock at that max speed they boost to. it's not sustainable.
I commented on your new watercooled build, you gotta get the active backplate for that 3090. It made a world of difference for me. A near 20c difference! 20c from a 90 degree air cooled temp.
Custom liquid cooling is such a small niche and a hassle even if price is no object. I probably wouldn't watch a video about a liquid-cooled backplate.
@@thedanyesful yet you watched a video about people replacing thermalpads on an air cooled 3090, which is significantly more niche than watercooling. Jay is also known for watercooling. I dont know what your deal is. But seems like you are out of your element, bud.
@@BamfhammerWow you really took that personally. Replacing thermal pads on a 3090 is maybe a $20 investment. Pretty big contrast to the cost of a liquid cooled build.
@@thedanyesful your argument literally included "if price was no object" and you are now citing price as a reason it is viable. Jay does watercooling and modding. People, like you, need to stop crapping over all ideas that might cost more than 50 bucks.
I wish I could add photos to this comment, I just did my RTX 2080 Strix because of this video. I swear there was almost zero thermal past on my gpu at all, hotspot was constantly at 105c and gpu temps holding at 81c while gaming. Did the thermal paste change. Install saw a hotspot of 79c while gaming and gpu stuck to 65c. Fans didn't constantly hold at 100% at 81c anymore. They didn't go over 40% after the thermal paste change. Massive difference, great video.
That would just be inferior to changing out the pads. Squishing cheap pads more will help them, but ultimately just using better pads are best. This is like saying add nitrous to an old 2003 Honda civic rather than just driving a faster car
@@daymianhogue1634 Add washers and pads, I have another comment mentioning pads. I use 2mm pads on my zotac 3080 along with washers to add mounting pressure which made a big difference ontop of adding better pads.
@@daymianhogue1634 And that analogy is terrible, by that logic it'd be better to buy a 3090 rather than try to improve temps on a 3080. You'd be buying a whole new car instead of trying to make a slow one faster. The 3080 still works just fine it's just a way to get the temps down a bit but by that logic instead just buy a whole new card that's faster.
I put an alphacool water block on my evga 3090 ftw3. When I took it apart, the original thermal paste was completely dry and had several missing spots on the die. Shocked at how bad it was.
yea and for mining you just needed to reduce power consumption and it reduced to heat enough for the cooler to maintain the memory at high performance. didn't even need to change the pads.
@@nickgordon7016 that's approach is fine with gddr6 memory, definetly not enought for gddr6X. Powerlimits affect the chip temps greatly, not the memory, especially in mining when you need to oc memory clocks.
okay, hear me out, if contact pressure is an issue, why not cool the VRMs with thermal compound? The only reason I can think of is cost but these are 3080s and 3090s we're talking about, I don't think cost would be an issue.
Thermal paste don't like to run hot, it dries out quicker, it's only used on the chip because it would be too much heat for a thermal pad to transfer, and it would get too hot (at least it's how I think it works, correct me if I'm wrong) also, the die needs to be as close as possible, VRMs would benefit too but not as much as the die
@@mke3053 even if you could, you'd have to change it every year at least, because as I said it dries out Also, Linus did a video on that, cooling vrms with thermal paste
After getting 110c temps on a 3090FE memory I used Thermalright Thermal Pads (1.5mm) which brought the temps down to 90c. The pads used on FE cards are practically insulators... The paste for the GPU was also bone dry, so I used a reputable paste which dropped temps of the GPU 15c.
Just did my 3090 fe with 1.5mm ThermalRight pads and Kryonaut paste. The temps were dropping significantly. People recommend getting 3 of the 85x45 size ones but I found out getting one 120x120 was enough and cheaper. Also adding some aluminum heatsinks to the backplate helps with a few degrees.
changing pads and paste on my Gigabyte 3080 made a MASSIVE difference. About 4-5C on the core and 20C+ on the memory, and that's after raising the memory overclock from +250 to +1300
No different than when I redid the pads and paste on my PS4 Pro last year with Thermal Grizzly. But at least in Sony's favor, the original pads and paste gave me three years of normal use first.
Everyone else: man I wish I could get my hands on a decently affordable upgrade GPU. This man: has enough to experiment on with no fear of losing multiple
I first noticed this issue when I started mining with my 3080 and the software showed the VRAM temp. Max mining, I had VRAM temps around 104-106 Celsius. Finished this mod an hour ago finally, running the miner ever since, fans at 80%, highest temp I've hit is 72!!!! 30 degrees celsius, WHAT?!
Not sure about the 30xx series considering I can't find/afford one, but applying a pad to the back side of the GPU die dropped my core temps by near 10C on a Strix 1080. Might be something worth testing on the 30xx cards
This ^ It’s scary to see how many comments are bashing JUST the FE models lol. Newsflash, it’s ALL cards. Asus, msi, evga, you name it. They use gddr6x, and shitty pads. Mod your card if you want longevity. Period.
I was wondering when i replaced the pads when i installed my copper waterblock how much of an improvement was just from using better pads and expensive paste. Gained 1200 points on firestrike vanilla just running the card bone stock. These companies are spending billions on chip manufacturing but cheap out on something as inexpensive pads . I guess they plan on the pads to deteriorate just in time for the new cards to come out so the normies will just buy a new card when their old one doesnt play the games anymore.
Great video. for the amount you pay for one of these, you'd think they would have sorted that out in manufacturing. you should do this on a AMD card too and see if there's room for improvement there too.
Would this be directly transferable to a 3080ti in terms of pad thickness? Just ran tests and my stock FE hovers around 94C memory and 74C overall, definitely some room for improvement I think.
Jay I don't know if you're going to see this but these are the correct measurements for the front side of the 3090 FE card. VRAM Front: 1.5mm (very soft pads like Gelid Extreme). 1.65mm is not required. Technically, if you want to be Mister Spock, it's left VRAM: 1.5mm, right VRAM: 1.6mm, but such a configuration is impossible with normal pads (Gelid Extreme is highly recommended due to softness) so 1.5mm works great. VRM Front: 2.0mm (soft pads like Gelid Extreme; left, right, single chip) Backplate VRAM: 1.5mm or 2mm pads (more choices here, EC360, Gelid Ultimate, etc). The exact measurements of the original Nvidia pads were taken with a calipers and were as follows: VRAM front: 1.5mm all VRM front left: 1.8mm VRM front right: 2.0mm (this is with "V" shape facing --->right) VRAM backplate: 1.0mm Backplate hotspots: not important. Seems to be a combination of 0.5mm (solid state caps, only saw this on a 3080 FE, not 3090 FE), 1.0mm and 1.5mm and 2mm, depending on location. The 1.0mm VRAM is a problem. Some people saw a 15C temp drop keeping the stock Nvidia pads (not opening the card) by using a custom 'bottom compression, e.g Lego" stand to make the heatsink fins "push"" into the backplate (case mounted card only). But one person found out if you remove this stand, the temps skyrocket on the VRAM to 110C, which means that the pads compress due to an air gap, but won't decompress, leaving temps worse. So the only way to fix this is to switch to thicker pads, then no compression stand is needed anymore. I use 2mm on the backplate everywhere (VRAM, PCB hotspots) then its great contact.
Can we maybe get a video where you test the difference between using copper shims and a thin thermal pad, vs just using a normal or thicker thermal pad? I haven't tried this method since my 8800gtx (I went through 3 of those cards), but iirc shims and thin thermal pads were supposed to be better than just using the thicker (stock or aftermarket) thermal pads.
The interfacing between materials is the weakest link of the heat transfer chain, so adding another layer ( two, really, since you'd likely need some kind of TIM between the copper and whatever the copper is touching ) adds more inefficiency. Memory chip -> thick pad -> heat spreader may not be great, but I suspect it'd take special circumstances for it to be worse than memory chip -> paste -> copper -> thin pad - > heat spreader.
Used Thermalright Odyssey 1.5mm on my 3090 FE and the VRAM temps when running port royal with 100% power is at 86C max, and with 114% power hits 88C max, all in a Lian Li case, with 9 of the Lian Li unifans granted up north in Michigan at like 70F room temp. the pads were cheaper than the Fujipoly, and are slightly better than the gelid (12W/K) at 12.8W/K.
I've added heatsinks and a small low profile fan to the top of my SAPPHIRE Radeon NITRO+ RX 5700 XT and my temps never get above 57 C. I haven't considered the memory pads though, but after looking at a teardown video, I see that Sapphire is using heat sinks on the memory. Great video though!
How insanely tone deaf is it for Nvidia to cheap out on something that can only cost them a few cents to do right? That's just Nvidia making it crystal clear what they really think of their customers.
Even in mass production the per-unit cost of a good thermal pad like Gelid or Fujipoly runs $20-40 for a sheet the size of trading card. Like the stuff that has similar heat transfer as thermal paste.
Seriously, look for thermal pads with high conductivity on Digikey or Mouser.
Using more expensive stuff would just increase the overall cost for the GPUs for not much benefit to Nvidia if it runs within spec.
Yes there is a significant difference with using better pads.
We’re talking about a company whose revenue is in the tens of billions every year. They’re all about the money.
@@rotor13 rather similar to how i make my sandwiches, i focus on the bread. just two slices put together and its ready to eat, no other ingrediants, sometimes i add a slither of butter as a treat.
Nvidia have focused on the pcb and the cooler and neglected the ingrediant that makes them work together when charging consumers enough money they could buy a car for. regardless of how much it would cost for efficient thermal pads, it comes across half-arsed for a crucial part imo. those temps are a joke. but i suppose less life expectancy is a good thing in todays world, if you'll need to buy another card sooner. products are no longer built to last. they just managed to meet their 'in spec' requirement, despite it being a poor result and costing a few grand to consumers. i would like a little more consideration for consistant quality throughout the product when i drop that type of money.
@@rotor13 for cards that sell for over $1000 a customer would expect it to be included in the price.
they were forced to use GDDR6X memory at fast clocks just to beat AMD's Infinity Cache with GDDR6 ram. Because of that, it makes the gpu(in particular the vram) run hotter just so they can get the halo card into the market to get mindshare in purchase. Cheapening out on the thermal pads is just whatever they had in stock. It's still the cheapest RTX card.
No doubt. Card with a $1500 MSRP (and good luck finding it for *that*) and they couldn't spring an extra $10 for proper cooling.
For the cost of the GPU, the paste and pads 'SHOULD' be set correctly. The fact this wasn't, doesn't worry me so much, but makes me think the manufacturer no longer cares about the consumer.
Pretty sure there is a target tolerance so that there is a fail rate so people will upgrade their cards earlier then necessary.
@@ricky4673 you could well be right.
Stuff is not made to last, it's made to cover the warranty period. After than, longetivity is actually a con for the manufacturer because selling more stuff gets harder.
@@MikkoRantalainen We are stripping the planet of valuable resources just to scam ourselves. Meanwhile the government keeps on talking about carbon bs that has little to do with the destruction of our planet or resources. We need to completely redesign our constitution as many new exploitative practices exist today that our forefathers never saw coming like medical tyranny, paid protestors, corrupted lobbying and severely corrupted political parasites.
@@ricky4673 For Nvidia it's pure and simple business and mass market bulk production (AMD as well for that matter). This means they will as often as possible try to get away with the least expenses as long as the main criteria for their core consumers are met. And they are. Them saving on pads when producing millions of cards, equals to big savings in the grand scheme.
What this also does, is ensure that their business partners like EVGA, ASUS, etc can make very minor and easy upgrades and still see significant improvements, which enables them to market their products more easily without having to spend excess resources on r&d for every new generation hardware. Changing pads and addind a few better fans can easily drop temps by 15-25 degrees on average, which means they can more easily increase the power limit for overclocking, and overclocking means they can see more benefit from using higher grade memory components etc. Basically Nvidia and AMD tries to make the upgrade path as simple and efficient as possible for their business partners. After all, these are the partners that buy the vast majority of the overall graphics cards (outside of stock pc retailers). Making their partners' lives easier is essential for Nvidia.
While this business approach has similar symptoms to something like planned obsolesence, it doesn't mean the outcome is the same. Most people who expect great performance and plan on maxing out their pc on a regular basis, are more likely to go for a 3rd party graphics card in the first place, like a Strix card which had the pads used in this vid.
This would have been a lot more useful if you actually gave a link or something to the pads that work well, rather than just saying "the grey squishy ones"
Once the plate is off and the chips/GPU are all cleaned, put a small amount of modelling clay on the chips and then put the backplate on. This will compress the clay to the exact thickness that exists between the components and plate. Carefully remove the plate and the compressed clay can be accurately measured. Pick the pad thickness based on that measurement.
Nice idea!
Don't forget you want about 25 - 50% compression on the thermal pads per spec.
@@RedHaloManiac95 yes, the modeling clay method is ok but it’s better to be on the thicker side rather than just perfect to provide pressure to provide maximum contact and thermal transfer and being under pressure the pads will eventually thin out slightly
Why not just put thermal paste on everything?
@@DescartesRenegade the gap is too large for thermal paste
This mod is mandatory IMO. When I opened my 3090 I have never seen such a poor job done at cooling from factory. Absolutely shocking. If you read around on this subject you'll find a guy who did some investigating and found that this memory starts to degrade at a much lower temperature than nvidia would have you believe. Pretty much every 3080/90 founders out there is slowly dying if you don't do this mod.
This entire series of cards, including the fucked up voltage regulation, really deserves an antitrust lawsuit or at minimum a class action.
Out of curiosity, does AMD have this problem with their Video Cards? I want to know because it would be good for influencing what I should buy in the future.
@@linear1224 Yes, they do. Memory temps on the reference cards are ass
cuz at nvidia's assembly lines, the staff are all Jays.
The cynic in me says they did this so miners wouldn't be able to sell second hand cards and scare people who buy used from miners when these cards start failing
gelid EXTREME 2mm pads work great for this, gelid 2mm ULTIMATE are too big, i think its because the extreme are squishy, i dropped around 20-25c on my 3080 FE from doing this mod
They sell in an 80x40x2.0mm configuration. Is one of those large enough to replace all the pads for a 3080 FE?
U got a 3080 ? I ain't got a gpu u some lucky man
@@rickynetherton2823 if you're careful yes, I used a 2mm and 3mm of those sizes and padded the front and back of it. I have a photo of my pads applied I could send if that would help guide!
@@T4nm4y Would love to see this picture too. Got some 2mm Gelid on the back of my 3080 FE and temps increased by 2-3 degrees…
Same temps with Frosty thermal pd 12.8 posted earlier above but saves your wallet. For FE i bought 1.5mm and dropped my temps identical to those pads. Works even better than Gelid Ultimate which I have no idea why. I heard Gelid Ultimate is too thick or doesn’t compress.
This is all well and good, but there's going to be a sizeable amount of people who just aren't technical enough to replace the pads. These GPUs should never have been released like this.
There's a sizeable amount of card's that will never have an issue too.
@@bepbep7418 not in a fairly regular lifecycle of 2-3 years. But that's not a good point to make. The cards running as hot as they do is a design flaw manufacturers can fix with literal cents.
@@noneofyourbusiness4294 but will customers buy new gen cards next year if their old cards work fine? especially today when you only need an upgrade if you're going from HD to 4K...Also it's not cents, buying a set of good pads could take $100 and more.
Exactly, it is totally unacceptable to sell cards like that, especially for what they cost. Modding should only be for those who want to exceed oem limits, it should never be necessary to do so, just to use the card under designed specs.
It's okay the one owner technical enough to do the job can just do it on the other 5 people who own 3080's as well.
nvidia: 'let's make most expensive gpu!'
also nvidia: 'let's use cheapest cooling thermal pads'
Not "make the most", its just charge the most.
@@JohnA... yup. btw do you know gaming gpu before 3090 which had more than 1.5k $ cost?
@@lukaogadze5071 Off the top of my head no, but maybe one of the Titan series? But honestly I don't think any have been quite that much before. At least not MSRP.
@@JohnA... Titan V and Titan RTX have higher price than 3090 but titan series were not created for gaming.
@@JohnA... 3090 was created for gaming but it's used for mining :D
Jay: Says that the test bench is in it's best case scenario.
Me: No that's a no case scenario.
This is the best comment.
But but, a no case IS the best case XD
Such a good bad joke lol
Lol
As Elon Musk would say, the best part is no part.
Imagine a company selling a 1000-3000$ high demand product and doing something as simple as setting up cooling on it correctly.
Wouldn't that be wild?
No shit. I sure hope they think about this on the new 4000 series.
Imagine people buying that poorly designed product.
It seems perfectly clear "they" want your card to die faster so you'll end up buying another sooner rather than later. Marketing tactics 101...
@@alpi8951 But that makes no sense when they sell out the instant they're in stock anyways. It's just penny pinching.
You've never bought a car? They cost even more.
Jay launchin a GPU benchmark : "As anyone know, the RTX 3000 series 3080s run very hot"
Jay after 10 min of GPU benchmark decides to touch the GPU : "Oh it's hot !"
This man will never change ^^
Knowing and experiencing are certainly two different things 🤔
I done the same when I first got a 980 Ti Classified a long while back. 80c on gpu, hotspot 92, vram/vrm around 100. I still touched the backplate though... and lemme tell ya, ya DONT wanna touch that thing when its running full bore.
@@deadly_mir figured that much
@@carsonm7292 Yeah every user loves to "experience" the warmth of the gpu. Nobody touches their GPU when the pc is on lmao.
@@timytimeerased I did. Many times for a couple days. Then I got bored of it.
"It was hitting 105 at his place, and it's like 78 in the warhouse right now." That casual mixing of temperature scales without so much as a pause :P
Shows you how remarkably easy it is for humans to understand both.
@@ericspda I for one have no clue how come they can survive so easily in 78C in the warehouse :D
Tbh I need to always convert with google from retard units
@@ericspda unless you live anywhere in the planet where farenheit isn't used at all
@@krzysiekj2522 hey that's pretty insulting to liberians
@@krzysiekj2522 I get it’s the cool thing to make fun of something you don’t understand, I just don’t see why people get so angry about a slightly different unit. It shows the same information with identical accuracy and typically higher resolution. Is it really that insane to consider a reasonable outdoor temperature as 75 degrees instead of 24? It’s all arbitrary anyways, it’s not like you frequently experience your body literally freezing or boiling, so it’s a bit confusing why water at standard atmospheric pressure is considered a gold reference everything should base off. Maybe you can educate me.
For anyone who wants to do this, make sure you actually measure the thermal pads using calipers to ensure you have the right size. some cards have 1 size pads, some have 2 different size pads, and some have 3 different size pads. you definitely want to make sure you have the correct size, saves you the trouble of having to re-open your card couple times and trying to figure out which pads might have been to big. just measure the the original pads, its better than guess work
@@WarAuthority my mitutoyo digital caliper be like :)
my 3080 has 5 different thicknesses 0.5, 1,1.5, 2 and 3mm basically every single thickness you can get is on my card, thank god I found someone with the exact same card that did all the hardwork of measuring for me.
Best results i know - one russian miner did change RTX3080 thermal pads to 1 mm copper plates and memory temperature go down from 100C to ~66C.
Not sure if it's his own idea but he did tried diffirent types of thermal pads/grease and all of them is nothing compare to copper.
Imagine if nvidia did this from the factory! This shouldn't be a necessary mod. Pretty frustrating
On a $1000+ GPU....
also warranty void if you fix our eff up
@@4kdragan439 Not in the US
@@NathanMc since when pretty sure as soon as you remove the pads unless you put everything back as it was you may not get your RMA accepted, depends on the party doing the RMA ofc but still that has been my experience in the past.
@@cr4zyw3ld3r In the us if they dont accept your RMA you just take them to small claims. They dont even show up to small claims since they know the laws not on their side so you win by default. Got MSI to refund me on two laptops that way.
Good work as usual. Manufacturers take the cheap way out with pads, replace!
Its sad how they cheap out on high end products that people pay so much only them to save so little
*WARRANTY VOID WARNING!, NVIDIA STATED REPLACING THE THERMAL PADS WILL RESULT IN ANY RMA/RETURN BEING DECLINED.*
Tens of people have already been told that doing this voids you warranty, so do it at your own risk. NVIDIA used crappy greasy cheap thermal pads and sadly if you replace them you see literally 15-20 degree decreases even. But you will also void your warranty. Just letting you all know!. Be safe!.
@@yakivpopavich that's pretty well-known fact like u take gpu apart ofc u gonna void warranty
@@yakivpopavich nobody cares
@@yakivpopavich That's a lawsuit to happen in a lot of countries.
Ugh, now I need to check mine. I really don't want to take it apart,
but damn you guys got some great results.
It’s worth it my 3090 was hitting 110C after like 10 min of time spy and now it tops out at 88C
I know i feel the same way. Will probably void my warranty.
Ok, so I ran Time spy and fur mark, but my memory junction only reached a max of 96C, of course I am running stock
out of the box EVGA FTW3 3080 TI. So it would seem as long as I leave at stock, I'm probably fine?
@@bowdiddly6172 yeah you are probably fine but you might be able to lower temps a little bit swapping pads and paste if you want.
This problem is mostly a big deal for the Founders Edition cards. I’m not really aware of widespread issues on any of the AIB cards
If you’re on a budget and use Amazon there is a strong product to check: Frosty thermal pad 12.8. If you are on a super budget i recently bought some 6 W/mk pads in amazon from Outus store or something . Its like $10 for 30 67x20 pieces. I am not sure about the Outis ones just yet but they claim to be 6 W/mk which is still way better than stock pads. Id use those for the backs and use Frosty for the front as they are 12.8 W/mk
I had the same issue with the pads and solved it with a pasta roller. Rolled my 2.0mm pads for a bit and stuck them where i needed to use 1.5mm. Worked like a charm
mamma mia !
@@32121 not even remotely joking, a card and some pressing also works....
They see me rollin' they hatiiin'
I used a piece of glass (technically a small mirror) to do the same, simply squished them down
Hard to get correct height, but sometimes its all you have left
I'd like to hear what Nvidia has to say about this. It's absurd this is happening.
The Thermalright Odyssey Extreme worked wonders for my 3080. Saw similar drops in the 20C range. Amazing that manufacturers skimp on $10 in better thermal pads for cards that are $800+.
Probably less $ bulk
It's on purpose. It's called planned obsolescence. The faster it overheats and breaks, the faster you'll buy a new one.
What size did you get for the 3080?
1:55 they are also "flip chip ram" so most of the heat is dumped away from the heat sink anyway... so thermal pads on the "tops" actually makes more of a difference than most think it will.
and the area is probably full of vias between the layers
*WARRANTY VOID WARNING!, NVIDIA STATED REPLACING THE THERMAL PADS WILL RESULT IN ANY RMA/RETURN BEING DECLINED.*
Tens of people have already been told that doing this voids you warranty, so do it at your own risk. NVIDIA used crappy greasy cheap thermal pads and sadly if you replace them you see literally 15-20 degree decreases even. But you will also void your warranty. Just letting you all know!. Be safe!.
@@yakivpopavich That depends on where you live, they can't void the warranty for opening up the card in the USA. That shit sucks balls for some countries to let them void the warranty just for opening it up and making it perform better and cooler.
@@yakivpopavich Yeah those stickers are not enforceable in Murica.
@@Apollo-Computers it makes sense because if the thermal pads potentially had something to do with the failure of the card Nvidia can't be expected to replace it because it isn't their fault that you modded it
I found that you need to first apply the pads to the card, NOT the cooler. The contact of the pads made best on the memory transfers to the cooler better. Not sure why...but lots of trial and error confirmed that every time I placed the pads on the memory first, it did better.
Thanks! This was really key. I was getting good results, but not what I expected, especially on Core temps. I just did this, with the same Gelid Extreme 2.0mm I used (opened and squished like 20 times in a row :( ) and it was GREAT. With undervolt to 1980@887 and +1000 on memories I was able to get almost 21k on air with the stock cooler. Paired with a 5900X on a skechy heatsink because I'm cleaning the liquid loop.
Rougher surface. Plus, one side is usually stickier (on purpose) than the other. And you want that side to be on the chips.
When you apply them with your finger, you can get more air bubbles out. Then when the heat sink goes on, it's usually flat enough that it's not an issue.
Used 1.5mm pads for the mod. Worked perfectly on my 3090FE. Be careful of the led cable.
You can simply measure the pads and if they are 0,8 you’ll need 1mm, if they’re 1,25 or so you’ll need 1,5mm because those pads where obviously squished a bit.
Also not too hard, not too much silicone, not too soft.
As for brands, Fujipoly 11 seem to be very good.
First thing I noticed with my Ventus OC 3080 was the insanely high VRAM temps. Bought some thermal pads and replaced only the pads around the die on the memory modules and mem junc. temp literally dropped by ~30C (from 108-112C to 80-82C)
What size?
PSA for the 3080 FE. The pads to use are the 2.0mm Gelid Extreme pads on the die side. On the back side you can use 3.0mm of Gelid or Thermalright. The reason why you have to use only Gelid Extreme is to Jay’s point, you really need 1.65mm but the Extremes are specifically soft so they can squish down to that size.
Yeah I used 2mm Thermalrights and I had to open it back up to squish them down more. I also applied extra paste to the gpu to help contact there. Gelid Extreme were out of stock at the time but they're definitely a key part of the mod.
@@ShimmeringSword yup I did that too got 3mm and 2mm and had to squish them down to get it to touch the plate. For some reason the noctua paste just couldn’t cool it good, so I got the kryonaut one and it helped. Temps don’t go past 82 on the vram. On a really hot day it got to 86, but that was at the peak of summer and not a well ventilated room. Definitely worth it. Got the gelid ones because I thought the thermalright ones were not going to work, but I ended up just keeping them and storing them away in case I gotta replace them down the road
I can’t believe how small the card itself is
Thats why its called a graphics card lol. The thing is 80 percent heat sink.
My Vega 56 is the same way lol I was so shocked when I unboxed it because the backplate only covers the PCB. The heatsink is almost literally twice as long!
I want to see some madlad make RTX 3080 ITX card, I know gigabyte makes a dual slot blower card 3090 that's only about 240mm long
@@t1995roger10 I know somebody made a little 3060, but temps should still be fine at 3080 power levels. My rule of thumb is 150W per slot at full length, as the 1070 katana is just a tad warm.
Die size is so small too. The amount of silicon is just dwarfed by cooling solutiom and power delivery
From what I have seen from other videos and browsing through reddit, it seems nvidia set the fan curve to memory temps rather than gpu die temps so if you do this mod and see gpu core temps rise then it is totally normal because the fans aren't spinning as fast due to cooler memory. So just create a new fan curve after you do the mod if you are worried about the gpu die temps.
I did this on my 3090 a few weeks ago with good results, vram temps dropped by a lot but my core temps were about the same or a little higher by maybe ~2C. I think that’s because the core can push harder now that the memory doesn’t throttle so much. My VRAM temps dropped by around 20C, which was a huge relief! The old paste they used on the GPU looked like it was 10 years old and the thermal pads were pretty bad so I swapped the pads for Thermalright Odyssey 1.5mm and I repasted the die with Kryonaut (and used way too much I’m sure). As always YMMV but it’s worth it given the awful stock pads and paste
I've heard the stock fan curve is based on memory temps, so since they're lower the fans aren't pushing as hard to cool the core
"It's within spec." Well, so is an Indy car that runs an incredible race and wins. It ran completely within spec...on the edge of its limits and is a million dollar paper weight in just 500 short miles. But, it ran perfectly within spec.
I like to run my ranger, C230, and Accord engines on the lower end of the spec. I be sitting there on the hottest day in summer with the A/C pulling the cabin temps to ice box.
The contently replace parts in the indie car. The start of the season car and the ending system could have had multiple components replaced to keep it in top racing condition
Dude I used Gelid pads on my PS3 on the backplates on the CPU and GPU it helped a lot. Those pads are great.
Did somebody say something??
@@drummerdoingstuff5020 no, its one of those porn channels.
"...Fortunately, I have one..."
Not gonna lie, I wanted to smack you. :P
Lol
OK Jay ... so on the 3090 you used 1.5mm and that worked. However, you then said you used "really thick pads on the backplate". How thick does "really thick" mean? Is it 1.5mm, 2mm, 3mm? What's "really thick"? (I don't use Facebook or Twitter so could you kindly please reply here). I need to know because I am ordering what I need (normally don't do DIY on computers and have no pads), since I am seeing the 100C temps on my 3090 and need to fix this.
I did this to both my 3090FE cards back in April/May and saw a massive reduction in VRAM temps. I used 1.5mm Thermalright 12.8w/mk pads. I even tried the stupid expensive Fujipoly 17w/mk pads, but the Thermalright pads were actually slightly better.
I read the the fibre/fabric pads than Nvidia uses are only rated for ~ 3 w/mk. That's shocking. £1400 each for the 3090 FE cards and they skimp on the thermal pads?
I drilled into the backplate of my 3090, drilled cavities into the backplate in between holes and ran 1mm heat pipes directly over ram module on the back plate, applied better thermal pads. Then i created a heat pipe array on top of the back plate and soldered a second copper backplate on top of that, then used thermal adhesive to glue on a bunch of aluminum heat sinks. Full memory overclock, +200mhz to GPU, 78C on ram after 1 hour of rendering with full ram usage!
sounds like something people would like to see a video or pics on dont it?
uh what
So they're charging all-time high prices for gpu's and they can't even put decent performing thermal pads and paste in them? Seems like they could spend the extra dollar to put nice pads in them
And you’d now have $50 more expensive GPU prices.
Good pads are expensive. Ampere runs HOT. They’re 300watt cards lol
@@rotor13 Fair point, but what's $50 on top of $1200 🤣🤣🤣
That's the ironic thing though, this time around the paste was really good. That made it even more infuriating to have to take my 3090 FE apart to replace the pads which got me a near 30 degree drop for the memory junction temperature. I used Arctic MX-5 paste for the core and it ended up around the same decent temperature of the stock paste, about 45C idle with "prefer maximum performance" on and only as high as 68C rendering and 71C gaming.
Thermalright Odyssey pads are godlike for the dollar!
I did thermalright pads for my 3080 and they are amazing!
@@gamester0123 1.5mm pads?
@@Mangokr 1.5 is what i did and i shaved even another 10 celsius off jays memory temps. can confirm the odyssey's are god tier
@@stanleybyrd7009 Thanks, I’m gonna getting those 👍🏻
@@Mangokr Make sure the measurements are EXACT. Odyssey's dont squish as much, so people have seen hotter temps as well because they arent high enough in certain spots, or too high.
Thats generally why people suggest the EXTREME as the squish factor is better. a 2mm squish will fit where a 1.6-1.7 is needed, where as 2mm Odyssey wont, and lead to higher temps/bad contact. Etc.
Edited. Not ultimate. Gelid Extreme. That is the one with better squish factor.
Having done this on my 3090, used 1.5 pads, massive improvement
Founders use 1.5 all around right or are they the nightmarish ones they use 1.5-2.0-3.0
What I did was add an "IHS" onto each memory package. The actual dies are absolutly tiny, about 3x3mm and the theory of installing an IHS is it spreads the heat out allowing more thermal pad to do its thing. If you go onto ebay, you can type in copper cpu shim and that will return thousands of results. With a little bit of cutting and filing the shims fit perfectly on top of a memory package. I never changed the stock pads on my GPU, just added shims and the temps dropped from 106c mining to 85c mining.
for those of you doing this mod......... try putting thermal PASTE on the memory modules and then putting the new higher quality thermal pads on. It made a huge difference on my card for the memory temps.
Jay i was so dissapointed, I bought my first ifixit kit this saturday and nothing exploded when i opened it up
It’s an iFixit kit, not Gigabyte! Lol
is this only for founder cards or does it effect partner cards like MSI, EVGA, and other card?
Different sizes of pads and vram temperatures for every variation. He was checking with software. One should too
Yes!
Like he showed in the video, strix had the good pads
Sometimes.
You need to check for yourself or google the specific card you have.
Near 100°C memory temps should be an obvious sign.
Did this on my Powercolor Red Devil 5600XT as memory temps were constantly 102-104C. Dropped to 80C! Same putty-like pads on stock.
Warning reader: I used 2mm and 3mm gelid ultimate on my card and my temps were worse. I tried a bunch of different variations too. The only thing that works and you dont have to mess with it over and over to get everything right; is using gelid extreme 1.5mm on the dye side. then u can fiddle with 2mm or 3mm on the back to find out what works best, i think the back its meant for 2mm though. its also extremely important that you get the gelid extreme pads exclusively, and not any other brand or style on the GPU side. gelid ultimates did not work better for me, in fact they made things worse.
On my 3090 FE I switched out the stock thermal pads for Thermalright 12.8 W/mK 1.5 mm pads and went from 110c max to 65c max on the memory.
If proper thermal pads drop the RAM temps by 20c, you can expect that to mean that the cooler is now handling more cooling load than it was originally intended for, so it is likely that the other components that need to be cooled will be cooled less efficiently. Probably a good idea to add an extra cooling fan blowing straight on to the card in that case (but be sure it's in the correct orientation so that it doesn't fight against the GPU fans!)
I would have liked to know, if the hot VRAM is only on the backside or ALSO on the front.
In other words: Can i improve VRAM Temps, just by changing the ones on the Backplate?
My reason for that assumption is the 3080 has only 1 VRAM Hotspot which can be cooled better by placing some Pads on the backplate.
no, you'll only cool half the memory, the other chips will remain just as hot
I got a gigabyte OC 3080 a while back and after putting on new thermal pads my temperatures dropped 15-20C
i just checked on my gigabyte gaming oc 3080, and its just fine. up to 84° on memory with 66% fan, at 100% down to 76°. also core and hotspot dropped 10° when going fron 66% to 100% fan.
@@od_static what's the ambient temperature in your room like? In an air conditioned space of about 72° f my gigabyte gaming OC was hitting about 90 95C on the memory modules with the fan at about 80 90%. Yes it's technically within spec but the cooler your card stays the longer it'll last.
Gigabyte is pretty known for using cheap leaky thermal pads. Yes I said leaky lol somehow their thermal pads leak oil after a while. Now under load I'm hitting around 75 85C on my memory and that's worst case scenario. It was also an excuse to put better paste on the core and that improved my temperature is another couple degrees too.
@@rocsoriano2994 currently 24-25°C Room temp. but i think i have to rerun with stock...sorry, i undervoltet to about 1995MHz@0.912V peak
@Lord Crydon What mm pads did you use? I also have a gigabyte 3080.
3080 gaming oc, had a look to under timespy and memory didn't go above 85 standard fan settings
My 3080Ti FE barely have any pads on the back. If you have a FE model, I highly recommended.
saw your past videos giving away rtx cards to your viewer's, thats awasome man, thats a very nice thing to do in this time of shorteges
i always get a mild anxiety attack when someone pulls apart a $5000 gpu
Can we get a product link for what you consider to be good thermal pads? 😊
I recommend Grizzly brand pads. They are expensive, but work extremely well and don't get destroyed when you lift up /off a few times
Fujipoly
I've had good luck with Gelid GC Extreme
*WARRANTY VOID WARNING!, NVIDIA STATED REPLACING THE THERMAL PADS WILL RESULT IN ANY RMA/RETURN BEING DECLINED.*
Tens of people have already been told that doing this voids you warranty, so do it at your own risk. NVIDIA used crappy greasy cheap thermal pads and sadly if you replace them you see literally 15-20 degree decreases even. But you will also void your warranty. Just letting you all know!. Be safe!.
@@yakivpopavich Good advice. The warranty is for 3 years which is really good for electronics.
Thank you so much for doing this video. I checked the temperatures on my Asus 3080 strix OC and is water cooled and was getting around 95C. Changed the thermal pads to thermal grizzly minus 8 pads and now the temps have dropped to around 65C. Huge improvement thank you
I wish Asus would once again have quality products... Any company that you can do a basic mod and see MASSIVE improvements is a company I would always second guess before considering purchasing any of their products :/
@@ihatesignupsgrrrrrrr I'm very happy to buy Asus products in the future as asus are not the only company this mod works really well on. At the end of the day the temperatures were within the spec of the memory.
I do a lot with software. Always nice to see these video's explaining the hardware side of computers.
I replaced the pads in my 3080FE with Gelid pads. Took memory temps from 104C to 82C. Completely worth it.
Hi, can you please let me know what thickness pads you used on both sides? And were they the Gelid Ultimates or Extremes? Thanks
Use my 3090 for mining and saw a 25C drop by upgrading the pads and adding some copper NVME heat sinks over the backplate. Have an old Gentle Typhoon A14 (?) blowing over them. Works like a champ.
Would love to see a follow up on this video with you using K5 pro viscous termal paste, instead of pads. Thanks for great content as always!
A review of aftermarket thermal pads would be a useful series. If you could fit it in??
Front and back side Gelid mods on the FE is a revelation. The stock pads are made of crumbly fail and frankly the stock die thermal paste application wasn't great either.
I watercooled my Gigabyte RTX 3080 with the Bykski waterblock with active backplate a few months ago. I used the blue thermal pads that came with the block. During mining I went from 110c(throttling/perf capping) with memory OC, to 90c. THEN I bought Gelid pads and just installed them this weekend, and I dropped down another 30c, to 60c with +1350 on memory. It's insane how much the thermal pads matter. I used 1,5 mm both on the front and back, for those wondering.
Ek pads are pretty bad due to their thermal conductivity in my experience. I recommend thermal pads by gelid. Also you could use thermal paste to fill in any gaps if you think the 1mm is too small and don't have any 1.5mm, or if the gap isn't exactly 1.5mm and is actually a bit larger.
I literally just did this to my 3080 Ventus and got exactly 20*C cooler mem temps and with only artic silver 5 paste got 8*C cooler core and I must have done a killer job with my paste because my hotspot is right on my core temp which was a 14*C drop! Also to note my ventus 3080 requires 2mm pads from the factory and I also had enough pad to replace the vrm and capacitors pads as well!
When is the washer mods coming?
Washer mods can lower temps again cause greater pressure.
Why not just park a car on your GPU?
Since i switched my 3090 Eagle to the Bykski block, i have changed the pads by default and they seem to be decent. Will test a bit more and see what i can get, i still have the hotspot numbers from the aircooler/stock pads.
Problem is there's a lot of conflicting information out there about whether this voids the warranty.
During these times where it's one per customer and I wouldn't be able to replace it myself, I'd value the warranty over the improved temps. You never know, your GPU could die of another cause beyond the crappy thermal pads and then you'd be out of luck...
Easy explanation of "but it's in spec": so is running a car engine at redline: that's the limit, you're running it at the limit. So... wanna try running your car to redline consistently over the life of your vehicle? Yeah, that's not a good idea. You can run it to redline, or even beyond under load, but it does cut into the life of your device.
I understand the underlying logic your analogy, however, it is flawed. You are comparing a mechanical engine with hundreds of points of mechanical failure, each of those parts with hundreds of points of failure and under multiple different amounts of forces and stresses.
GPU mining is something everyone hates, but, as multiple outlets have covered and shown, has no tangible effect on the card. GamersNexus showed that a card mining at 100% utilization for ~4 years straight versus one that was basically new in box, were performing identically to each other. Will the GPU one wear out before the other? Seems logical to assume, but there is zero evidence to suggest it to be so.
Electronics are robust and the stresses they experience aren't the same as mechanical stresses. Running a normal consumer car at redline for say 10 minutes straight will blow the engine, if I run my GPU at 100% utilization for 10 minutes and it fails, it isn't my fault.
@@johnlahey2157 100% utilization and reaching the max thermal capacity limit are not the same thing.
if i run my sound equipment over 75% it's not sustainable and will overheat.
similar to "boost" clock on cpu's. there's a reason why they don't just clock at that max speed they boost to.
it's not sustainable.
Yep.... replaced the pads on my FE, same result, 105C to 80C on memory. Crazy how we have to mod it like that for it to function properly
I commented on your new watercooled build, you gotta get the active backplate for that 3090. It made a world of difference for me. A near 20c difference! 20c from a 90 degree air cooled temp.
Custom liquid cooling is such a small niche and a hassle even if price is no object. I probably wouldn't watch a video about a liquid-cooled backplate.
@@thedanyesful yet you watched a video about people replacing thermalpads on an air cooled 3090, which is significantly more niche than watercooling.
Jay is also known for watercooling.
I dont know what your deal is. But seems like you are out of your element, bud.
@@BamfhammerWow you really took that personally. Replacing thermal pads on a 3090 is maybe a $20 investment. Pretty big contrast to the cost of a liquid cooled build.
@@thedanyesful your argument literally included "if price was no object" and you are now citing price as a reason it is viable.
Jay does watercooling and modding. People, like you, need to stop crapping over all ideas that might cost more than 50 bucks.
I bet you more people modded their thermal pads on 3090s and 3080s than have literally EVER built a custom liquid cooled system in all time.
i'm glad the heatsink has spots that show where to apply the pads, when I have to eventually do this fix on my 3080ti it should help a little
On the memory chips. Are you kidding
I wish I could add photos to this comment, I just did my RTX 2080 Strix because of this video. I swear there was almost zero thermal past on my gpu at all, hotspot was constantly at 105c and gpu temps holding at 81c while gaming.
Did the thermal paste change. Install saw a hotspot of 79c while gaming and gpu stuck to 65c.
Fans didn't constantly hold at 100% at 81c anymore. They didn't go over 40% after the thermal paste change. Massive difference, great video.
Adding very thin washers to the mounting screws helps a ton try that out.
That would just be inferior to changing out the pads. Squishing cheap pads more will help them, but ultimately just using better pads are best.
This is like saying add nitrous to an old 2003 Honda civic rather than just driving a faster car
@@daymianhogue1634 Add washers and pads, I have another comment mentioning pads. I use 2mm pads on my zotac 3080 along with washers to add mounting pressure which made a big difference ontop of adding better pads.
@@pottasiumn3231 ah my bad, this comment of yours is the only one I saw and it doesn't mention pads at all.
@@daymianhogue1634 And that analogy is terrible, by that logic it'd be better to buy a 3090 rather than try to improve temps on a 3080. You'd be buying a whole new car instead of trying to make a slow one faster. The 3080 still works just fine it's just a way to get the temps down a bit but by that logic instead just buy a whole new card that's faster.
@@daymianhogue1634 It was just something else I've tried and worked well so figured I'd mention it aswell.
That "i got another one (3080)" flex.
Made me laugh and think about everything and i achieved tranquility.
I used the giled ultimate 1.5 mm for my 3090. I put them on the front and back of the pcb and I got ~ 20 ° C cooler vram temps.
How many cm² (or whatever unit they're measured in) do you need for such a mod?
I guess warranty is voided, if you do such a mod?
squishy ones ? can you put some link where to buy ?
I put an alphacool water block on my evga 3090 ftw3. When I took it apart, the original thermal paste was completely dry and had several missing spots on the die. Shocked at how bad it was.
11:34 "STEALING THEM FROM A GRAPHICS CARD YOU PERHAPS DON'T NEED ANYMORE" - What alternative reality is this guy living in? Elitist or what!
Imagine cheaping out on (current price here if you want one) 1400 and 2200€ gpu's. Insane.
The mining community did discover this gddr6x issue a long time ago, surprised that jay just founds this out 😁
yea and for mining you just needed to reduce power consumption and it reduced to heat enough for the cooler to maintain the memory at high performance. didn't even need to change the pads.
@@nickgordon7016 that's approach is fine with gddr6 memory, definetly not enought for gddr6X. Powerlimits affect the chip temps greatly, not the memory, especially in mining when you need to oc memory clocks.
okay, hear me out, if contact pressure is an issue, why not cool the VRMs with thermal compound? The only reason I can think of is cost but these are 3080s and 3090s we're talking about, I don't think cost would be an issue.
Thermal paste don't like to run hot, it dries out quicker, it's only used on the chip because it would be too much heat for a thermal pad to transfer, and it would get too hot (at least it's how I think it works, correct me if I'm wrong) also, the die needs to be as close as possible, VRMs would benefit too but not as much as the die
@@lorenzocampolucci4464 actually paste is better, but cant have the high you need. Pads are for when there is need for more than paste high.
@@mke3053 even if you could, you'd have to change it every year at least, because as I said it dries out
Also, Linus did a video on that, cooling vrms with thermal paste
After getting 110c temps on a 3090FE memory I used Thermalright Thermal Pads (1.5mm) which brought the temps down to 90c. The pads used on FE cards are practically insulators... The paste for the GPU was also bone dry, so I used a reputable paste which dropped temps of the GPU 15c.
Just did my 3090 fe with 1.5mm ThermalRight pads and Kryonaut paste. The temps were dropping significantly. People recommend getting 3 of the 85x45 size ones but I found out getting one 120x120 was enough and cheaper. Also adding some aluminum heatsinks to the backplate helps with a few degrees.
What do you mean by "adding aluminum heatsinks?" Which ones did you use?
changing pads and paste on my Gigabyte 3080 made a MASSIVE difference. About 4-5C on the core and 20C+ on the memory, and that's after raising the memory overclock from +250 to +1300
No different than when I redid the pads and paste on my PS4 Pro last year with Thermal Grizzly. But at least in Sony's favor, the original pads and paste gave me three years of normal use first.
Everyone else: man I wish I could get my hands on a decently affordable upgrade GPU.
This man: has enough to experiment on with no fear of losing multiple
for clairification, this isn't a problem for the 3070s?
Which pads did you use for the back plate ? 1, 1.5 or 2?
I first noticed this issue when I started mining with my 3080 and the software showed the VRAM temp. Max mining, I had VRAM temps around 104-106 Celsius.
Finished this mod an hour ago finally, running the miner ever since, fans at 80%, highest temp I've hit is 72!!!! 30 degrees celsius, WHAT?!
Fuck people who mine with these cards
It’s generally pretty disappointing how power and temps have worked out with these cards
Thank you for this, Jay!
Please check out AMD cards too! I wanna know if there are thermal issues for the 67-6900 series!
Not sure about the 30xx series considering I can't find/afford one, but applying a pad to the back side of the GPU die dropped my core temps by near 10C on a Strix 1080.
Might be something worth testing on the 30xx cards
Is this a common issue with the lower cards like the 3070ti/3070 as well?
how would you know though? they don't have sensors for memory temperature.
I was 1 year faster than JayzTwoCents on something lol. Every single 3080 needs replacement or adding of thermal pads.
This ^
It’s scary to see how many comments are bashing JUST the FE models lol.
Newsflash, it’s ALL cards. Asus, msi, evga, you name it. They use gddr6x, and shitty pads.
Mod your card if you want longevity. Period.
I was wondering when i replaced the pads when i installed my copper waterblock how much of an improvement was just from using better pads and expensive paste. Gained 1200 points on firestrike vanilla just running the card bone stock. These companies are spending billions on chip manufacturing but cheap out on something as inexpensive pads . I guess they plan on the pads to deteriorate just in time for the new cards to come out so the normies will just buy a new card when their old one doesnt play the games anymore.
Great video. for the amount you pay for one of these, you'd think they would have sorted that out in manufacturing. you should do this on a AMD card too and see if there's room for improvement there too.
I think it comes out to the board manufacturer, not just the chip.
Would this be directly transferable to a 3080ti in terms of pad thickness? Just ran tests and my stock FE hovers around 94C memory and 74C overall, definitely some room for improvement I think.
What size did you use?
Jay I don't know if you're going to see this but these are the correct measurements for the front side of the 3090 FE card.
VRAM Front: 1.5mm (very soft pads like Gelid Extreme). 1.65mm is not required. Technically, if you want to be Mister Spock, it's left VRAM: 1.5mm, right VRAM: 1.6mm, but such a configuration is impossible with normal pads (Gelid Extreme is highly recommended due to softness) so 1.5mm works great.
VRM Front: 2.0mm (soft pads like Gelid Extreme; left, right, single chip)
Backplate VRAM: 1.5mm or 2mm pads (more choices here, EC360, Gelid Ultimate, etc).
The exact measurements of the original Nvidia pads were taken with a calipers and were as follows:
VRAM front: 1.5mm all
VRM front left: 1.8mm
VRM front right: 2.0mm (this is with "V" shape facing --->right)
VRAM backplate: 1.0mm
Backplate hotspots: not important. Seems to be a combination of 0.5mm (solid state caps, only saw this on a 3080 FE, not 3090 FE), 1.0mm and 1.5mm and 2mm, depending on location.
The 1.0mm VRAM is a problem. Some people saw a 15C temp drop keeping the stock Nvidia pads (not opening the card) by using a custom 'bottom compression, e.g Lego" stand to make the heatsink fins "push"" into the backplate (case mounted card only). But one person found out if you remove this stand, the temps skyrocket on the VRAM to 110C, which means that the pads compress due to an air gap, but won't decompress, leaving temps worse. So the only way to fix this is to switch to thicker pads, then no compression stand is needed anymore. I use 2mm on the backplate everywhere (VRAM, PCB hotspots) then its great contact.
I had a guy run into this issue on day one of enjoying his 2 new 30 series, the thermal pads that sit on the memory were completely misplaced.
Can we maybe get a video where you test the difference between using copper shims and a thin thermal pad, vs just using a normal or thicker thermal pad?
I haven't tried this method since my 8800gtx (I went through 3 of those cards), but iirc shims and thin thermal pads were supposed to be better than just using the thicker (stock or aftermarket) thermal pads.
The interfacing between materials is the weakest link of the heat transfer chain, so adding another layer ( two, really, since you'd likely need some kind of TIM between the copper and whatever the copper is touching ) adds more inefficiency. Memory chip -> thick pad -> heat spreader may not be great, but I suspect it'd take special circumstances for it to be worse than memory chip -> paste -> copper -> thin pad - > heat spreader.
I did this on my 3090FE and it brought my temps down by over 30C
Used Thermalright Odyssey 1.5mm on my 3090 FE and the VRAM temps when running port royal with 100% power is at 86C max, and with 114% power hits 88C max, all in a Lian Li case, with 9 of the Lian Li unifans granted up north in Michigan at like 70F room temp. the pads were cheaper than the Fujipoly, and are slightly better than the gelid (12W/K) at 12.8W/K.
I've added heatsinks and a small low profile fan to the top of my SAPPHIRE Radeon NITRO+ RX 5700 XT and my temps never get above 57 C. I haven't considered the memory pads though, but after looking at a teardown video, I see that Sapphire is using heat sinks on the memory. Great video though!