One thing I want to add: the previous owner used his card vertically! Because a lot are drawing quick conclusions that "it's only because it was not used horizontally". It doesn't seem to be this simple :) I organized two more completely stock cards today and will do a follow up with more testing soon - including different positions.
@@royboysoyboy I hear you there but I'm thinking more along the lines of what happened to the 3090's and new world. Just hammering one part of the card in that case to failure, but here the card is keeping its self in "acceptable" limits.
You must test different positions; go look at all the Reddit posts - people report 110c in one position and then 80c in a different position, so the orientation matters
I guess thats fine, but i have no reference to kelvin, since i don't use it. If he said 25celcius, i would know right away without thinking, but when he says 25kelvin, i have to check what it means.
The problem was completely missed in the video. I have tested my reference rx 7900 xtx and have observed that following: Traditional orientation (horizontal with fans facing down) yields hotspot temperature of 110 C Vertical mount (like shown in the video) the hotspot temps range between 80 C to 90 C After testing and verifying these results, I performed an additional test where I took off the backplate to tighten all the mounting screws. Most of the screws were already sufficiently tight, but some had a tiny bit more turning capacity by hand. The results after hand tightening the screws were the same, with the horizontal mounting orientation yielding major temperature issues. The mounting orientation yielding such drastically different results means that the problem is either with bad mounting pressure on some units or bad vapor chamber on some units. I would love to see more thorough investigations as to why mounting orientation is causing this issue and what can be done to fix this issue, other than vertically mounting the GPU, which is not possible in most PC cases without modification. Please consider these recommendations and I really hope to see another video update from you regarding this topic.
I tested my card exactly the way the previous owner used it :) he did not have horizontal mounting. But I will look into this. I don't think it'd this simple tho
@@der8auer-enI have this card, I had major hotspot issues when the card was mounted vertically (not traditional vertical though, I had mine with IO pointing to the floor). I then changed to traditional vertical mount and the temps were then ok.
Mine was vertically mounted in my meshlicious and had the issue. I did not bother to repaste tho. I just returned it. Laughable that so many cards have this issue from the factory.
@@der8auer-en Could it be possible some coating in vapor chamber got loose and it's moving around, blocking something and after taking card and moving it around it unblocked?
I have the reference 7900 XTX mounted horizontally. With stock paste and mount I was at about 15 to 20c delta, maxing in the low 80s for tjunction. I decided to repaste for fun with kryonaut (not extreme) and it maybe lowered both temps by 1 to 2c, so the stock paste and mount was actually perfectly fine.
For me, the issue only becomes visible once I unlock the powerlimit from 350 to 400W. Before that, the delta is quite big (30°C) with junction in the 90s but after that, it can go from 60-70°C edge temps to nearly 50°C delta with 110°C junction temps that cause throttling. This is a card in Hamburg, Germany. Could get that one to you - but I think it'd be better to find a card that shows a bigger delta out of the box.
as I explained in a comment I had the same as you with a pre-waterblocked 4090 factory I had 61-112 !!! gave my warranty up and opened it, a good 3rd of the gpu position on the block was clear, wasn't touching, cleaned repasted with thermal grizzly kyronaut and it's now 46-56°C for 440w I doubt I'm the only one who got a bad contact/paste videocard
@@fredEVOIX Just opening a GPU is not modification, and will not void your warranty alone. Granted, AIB's will always err on the side of caution and say something along the lines of, "retention bracket screws scratched." That is still not sufficient evidence enough to void a warranty under Moss Act, if residing in the US. Small claims court filing fee is usually few hundred dollars, small price to pay if you're replacing a $1600 card.
@Jonas Jonaitis Except, it's economically not small whatsoever, which is the context here. Europe has even better laws to protect the consumer. So what is your point?
Try to mount it horizontally like in a normal case, thats when most people get issues with hotspots. Vertical mounting gets rid of it from what I've heard.
This. Someone on reddit claims the temps are normal if he lays his PC case on its side, but if his case is upright the problem returns. I imagine der8auer is using a test bench instead of a normal PC chasis, which might explain why he didn't have the issue his friend did with the card.
This is absolutely true. I have tested and verified horizontal vs vertical mounting orientation and observed that the problem goes away when mounted vertically.
@@der8auer-en Remember that screwing down the card's metal bracket to a case can change the tension distribution across pcb and cooler. You have to take this into account in further analyses.
Would be interesting to see if the orientation of the card would change the result. Vertical mount (as tested here) vs. horizontal mount (as it would usually be mounted in a case).
I agree, it could be a horizontal bending issue. And there could be a high temperature in the user's case. I didn't see the GPU temperature from the user when the hot spot comes to 109 degrees.
If 7900xtx has the anti-gravity heat pipes then the difference should be a few degrees C, if not probably >20 C. But it shouldn't change the delta too much between GPU vs GPU hotspot
I had rear some reports about temperature problems when mounted horizontally that went away with vertical mount, so if roman uses only testbench he wont get same results as if it was in case.
My reference card had 110° hotspot and 60° gpu temp. 3000rpm (100%) fan and downclocking. I just returned it. I'm perfectly fine with changing thermal paste but I'm not doing it with a brand new card of this cost. Completely ridiculous.
I overclocked my 4080 +150 on the core and +600 on the memory (founder's edition) The GPU temp doesn't reach more than low 60's and the memory hasn't hit 70c yet.
@MultiNastyNate exactly. Have a 4080 now to. Fans don't even turn on playing warzone 2 at 1440p 200+fps lol. Sits at 65° 1300rpm fan in games that max usage.
There was some talk on reddit about the possible issue being people using a non certified display port cable causing the issue, related to the 20pin in the cable. This could be one of the variables as to why Der8auer could not replicate the issue
The orientation of GPU is important as well, if its a test bed, then most probably you wont see a problem, people reported that tilting the case to its side like a test bed results in lower junction temp, so if you are doing it on a test bed, then just try installing the card horizontally (fans facing down) and the case sitting vertically like a normal PC and then retest on new affected card since the one you have already got fixed by re-applying thermal paste and balancing the pressure.
I was also thinking about whether it was in a case or on a test bench, which is usually -- if not always -- open-air. That, combined with the orientation, will definitely impact the air flow.
I personally owned a single fan gigabyte gtx 970, it was running 60C in the mobo and 88C in the vertical GPU bracket, that was when i found out about this issue, it was easily reproducable by tilting the case on its back or face. Vertical bad temp horizontal good temp
I’ve got data on this. Orientation creates a higher temp delta. It’s crazy. Test bench style mounting is okay Normal case mounting causes 110 Celsius junction temp.
You people keep saying orientation matters but Roman already commented in this comment section saying he has tried both orientations and it is irrelevant.
Yes, that was one of the theories was that mounting pressure was probably the most common cause as they were rushing to get enough stock out for launch. But it is still great to have this confirmed by someone like yourself. Thank you. (I have one and no issues what so ever. Even OC and u-v very well).
@@acidgaming0 sorry to hear that. I wish we had a rough percentage of product with issues. I guess Powercolor is collecting this data for AMD. Very sad indeed. I hope this gets resolved and you get looked after asap.
@@ktpt3732 It does sound that way. Some of us very happy with their product, others, like acidgaming0 above, received a bad model. Very sad. I hope this gets resolved and these people get looked after asap.
My XFX 7900XTX reference (horizontal) would reach 110C hotspot within 6 minutes~ of running furmark. I did open it up to repaste with MX4 and did torque the screws as far as they would go by hand but unfortunately my temperature issue persisted. I did test vertical mount by flipping my case over and it did definitely have dramatically better results. At this point I don't know what to do besides a RMA since I plan to use the gpu horizontally. Also when repasting I noticed the stock thermal paste was very solid, best way I would describe the consistency is very old and worn out thermal paste.
My 5700xt strix displayed similar temps, turned out the thermal pad on the fets was a whole 1mm to thin so the screw heads made no contact to the back of the pcb, simply folded the pad and it dropped junction temps from around 108 to high 80’s. Also the die was pretty scratched from the cooler sitting crooked for so long and possible having a loose cooler when in shipping.
@@SianaGearz I was told by the ASUS rep it was AMD’s fault for sending out faulty measurement’s and I could just send it in but I wanted to see if I could fix it first or figure out what they did wrong but yeah so much for q&a the screw heads were literally a whole 1mm from the back of the pcb on VRM side.. lol
I could guess the issue might come from the squishy pads on the memory chips that could possibly affect the pressure on the GPU die, now when they been opened and messed around with they probably been flattened more and the pressure on the die probably increased again
@@christophervanzetta if the pads on the AMD cards would be the problem it could be so easy as to just apply pressure around this area by hand to push the card down to a flush level. Unless the paste get dried up from the high temperature very quickly
Thanks for your efforts on making this video. I have been experiencing those issues with my recently bought 7900xtx sapphire reference mode. At first i thought, that i was installing in the wrong way my driver or my case's fans weren't working properly. I hope with videos like this, even that is going to cost a lot, AMD should accept RMA's and learn from this experience. And don't cheap out on the thermal paste in a 1k USD/EUR GPU.
This is unfortunate because the cooling solution including the paste used on the die is superior to what they used in the previous gen (check GamersNexus's teardown review); however, you can have all sorts of problems in the assembly line. These should be caught in QA test but obviously, they weren't.
So i had exact same problem with my 6700XT. Weirdly enough problem was solved not by thermal paste but by adding support bracket to a card. It somehow bent itself to the point that part of GPU lost contact with a cooler i guess.
what exact card do you have? I recently bought MSI MECH 6700XT and same issue here. Just asking because maybe I can fix mine too without sending it for RMA.
When you RMA a GPU and they test it, they will most likely position the GPU vertically as the testing board they use it on will be flat on a table... This also means they don't notice any problems that you would see if the GPU is installed horizontally in a case.
I’ve just put my Powercolor 7900 XTC MBA in my Pc and ran kombustor. After 1 minute my junction temp hits 110 Celsius and the GPU is at 59 Celsius. Even with fans at 100% it takes 2 minutes to get to the 110 Celsius junction temperature!
On the AMD forums (I can't provide actual links because UA-cam deletes my posts if I do), one person tested his card in four orientations (look for "7900xtx reference orientation testing. Orientation impacts performance"). The bad temperatures show up when the card is a "normal" orientation (i.e., when the card is horizontal). In a "test-bed" orientation (when the card is vertical as with your setup), the temperatures are fine. Try tipping your test-bed on its side or putting the card in a regular, tower-type case.
I was re-pasting my 2 y old 6900xt yesterday because hot spot was reaching 105* C but GPU temp was never above 64* C after doing this the hot spot temp stops at 82* C but the GPU temp goes up to 71* C I was surprised because the original paste looked good and was still quite soft
@@DigitalJedi Nah it's because the card before repaste hit the tjmax at hotspot so it would clock down. Higher temperature in this case means more performance due to higher amount of workload being performed.
High hotspot temp could mean the chip is not fully covered in thermal paste. Bad news if the die has no heat spreader. Heard of this happening on nvidia gpus as well and that it affects some amd 6000 series as well.
I encountered this issue with my reference 6800XT when it was vertically mounted with ports up - hotspot temps would be through the roof and half the radiator would be cool to touch.
Thanks for this video. I just stress tested my XTX. Results here as a point of data for others: 65 degrees F in room GPU temp never rose above 65 C and junction temp which has been the hotspot on some of these cards never rose above 79 C even under sustained 100% load. The fans never went above 62% to maintain those numbers.
I have reference XTX bought from AMD's site in US. My junction temp rarely reaches 90°. It seems that the hotspot problem is more common with models sold in EU. Anecdotal evidence only, but I've certainly seen more EU users complain about the hotspot temp problem.
air pocket? I've noticed it near impossible to see signs of one after you pull the heatsink as it suctions away any traces, unless the thermal paste is old enough to leave a bubble mark. I always make an 'X' with a little extra in the middle to ensure the air gets pushed out. With Liquid Metal i put an even coat just enough to cover the surfaces with surface tension then a little extra on the middle of the GPU/CPU and the middle of the heatsink. Over time the air pockets might pop and temps will go to normal, but that can take weeks of temperature fluctuation to work itself out
Is GPU sag from horizontal mounting vs vertical mounting a factor? Also were the screws possibly not torqued enough? It seems like either disassembling and reassembling the card or mounting it vertically corrected the problem.
I haven't heard of it being a problem, those screws are made to be torqued until they hit a stop and not force further, the torque is given by the metal bracket (spring). The card can't be easily influenced by sagging as it's got many screws which keep it from flexing, especially with the help of the backplate. Most of the visible sagging would come from the PCIE slot itself plus a little bit from the edge of the PCB which inserts into the PCIE slot, but that little bending there is just local and doesn't reach the GPU area. In this case, I would just assume the owner either mounted it incorrectly or used a bad TIM or didn't clean the chip well enough before reapplying TIM. Also, the issue might not manifest with normal PL in place, might need to give it maximum power to see bigger Deltas.
9:32 maybe it only happens in certain games like Call of Duty MW 2. That game runs abnormally well on these GPUs. Like beating a 4090. Maybe it's using silicon that other games aren't using. Maybe Furmark is not activating whatever is causing this.
@@humorss Well there are, there's the shader cores with separate ALUs for different kinds of arithmetic operations, and of course even in one multipurpose ALU, all operations don't hit the same exact semiconductors, and around them there's TMU (texture fetch), memory/scheduling (because memory fetch often forces latency hiding logic to swap tasks), and ROP. And that not even counting the new RT hardware. When you read utilisation, it just takes the max of all values per SM, and any operation in an ALU is considered as that ALU being utilised, in spite of some being potentially more power hungry than others. Typically, ROPs are severely underutilised in most of the render passes.
My reference card is also getting 110 Celsius hotspot temperature. I've tried already new thermal paste, used different bracket pressure on GPU by adjusting GPU brackets crews. No difference so far. I'm waiting what you will discover. I'm curious what is your method of placing bracket on the card. Which pressure did you used. Thanks for looking into that problem. Can't wait for that video.
Hello. I also have a 7900XTX XFX Merc 310 card, the hot spot cannot even be cooled with a water cooler. The temperature of the hot spot was still around 110 degrees and I did what I could (various thermally conductive pastes, tapes, etc.). I'm quite stressed about it, because the card doesn't cost a lot of money. I would also like to point out that there are of course different temperatures in full HD, 2K and 4K at certain Hz, where 60Hz will have low temperatures. Where, of course, the high temperatures are resolved at 144 Hz in 4K! That's pretty important!
One thing that many people think is that in a vertical mount in a test bench it’s fine but mounting it horizontally in a pc the cooler pulls away from the pcb and die slightly. I need to finish the video before I comment i see you talked about this issue above but i will watch it all love your content.
Thanks for this, I had already repasted my RX 7900 XTX and it did help a little (96c junction temps at stock) but I just did so again with kryonaut extreme and also made sure tigten everything down correctly so far at stock settings max junction temp is 86c, biggest delta I've seen is 21c!!!! Yay!!!!!
How long was the card stressed? If they're using phase change material like the honeywell pads, temps will be higher than normal for around the first 24 hours of use.
@@der8auer-en hi, I’m retesting after running stock settings on my card and in time spy extreme are hitting about 90 degrees on junction and 61 on the gpu temps just wondering if this sounds about right, closed case about 25 degrees room temps
Try to run card horizontally mounted. Seems that some cards has some vapor chamber issues and cards works better when verically mounted (as they are in test bench etc...).
Here's an update, I have a 7900XTX originally from Sapphire, but it's mounted with a Bykski waterblock. The GPU works at around 45 degrees, while the GPU Hotspot reaches 110 degrees. I'm already using the GPU vertically and I don't know what the problem is. Is anyone else in the same situation?
Honestly that's the best case scenario. If anything AMD can send out some free thermal paste and post a video about how to disassemble the card and properly re-paste it. If people don't want to wait they can say you're welcome to re-apply paste on your own and we won't void the warranty. Not nearly as bad as the difficult to clip in 12/16 pin connectors.
Some people bought new PSU where 12 pin to two 8 pin connectors are given to them... i wonder if the 12 pin slots on the PSU outputs different wattage than normal 8 pin slots.
This card has some serious issues. Mine was running hot at 110 junc with a bad dp cable. Swapped cable - temps went down 20°. Tore the PC apart few days later. Again - 110°.. Guess what? Different temps depending on which displayport port you connect to. I can imagine that you didn't fix anything but rather don't encounter the problem because of some fluke. Interaction with display or whatsnot
my powercolor card will run at 397w for 2-3 mins, then the heat starts to overwhelm the card, fans go up to 2900 rpm, sounds like a vacuum cleaner ALL DAY LONG... wattage will lower to 330w. Horizontal mount is about 10° hotter than vertical mount, hotspots temps are 110° no matter what. Boxing it up and taking it back.
I have a reference asus 7900xtx and have had 40*c delta's in cyberpunk. I tried reproducing the issue in furmark but only saw a max temp of 90*C in furmark. In cyberpunk however after about 20-30 minuets of gaming my card had a 40*c delta and junction temp hit 110*'c with gpu temp at 70*C, when junction temp hit this temperature the card went from a 2500mhz clockspeed to a 2300mhz clockspeed and fans hit 3000 rpm. my card has standard horizontal mounting.
Thanks for this. I have a Sapphire Toxic 6900XT I bought this past July. It ran great OTB gpu temps in the 60's while gaming in the Division 2, hotspot temps in the low 80's. But within 3wks temps went up 10C on the gpu and hotspot was in the high 90's. After a week of temps hitting 100C, I took it apart and repaste the gpu with Kryonaut, temps dropped back to what I saw in the beginning. But yet again 2-3 wks later temps rose back to high 90's on the hotspot. I repasted again but with IC Diamond. It had appeared to me that the Kryonaut had been squeezed out. IC Diamond is a thick TIM. This fixed the issue for the last 3 months. I didn't have this issue with my cpu and Kryonaut. I guess it's just luck of the draw.
IC Diamond shouldn't really be used for GPU's or CPU direct die. As GPU's don't have a heatspreader and you are putting the paste directly onto the die it self. IC Diamond is just that, it has diamond particles inside it meaning it can scratch the die surface and MAYBE damage it over time.
@@Fezzy976 You show your ignorance with that statement. Maybe you should actually look it up on their website before you make ignorant statements of Bull. Nobody beats IC Diamond for thermal conductivity and it lasts for years. It doesn't get squeezed out by thermal expansion and contraction like thinner pastes.
@@jerrywatson1958 How am I being ignorant? I was just stating a possible outcome based on facts. This is the same kind of reason for not using Arctic Silver 5 or other metal based TIMs on direct die due to them being electrically conductive. Sure it probably won't fry your harware if you do it right but there is always a chance of human error. There is so little difference between most of the high end thermal pastes that thermal conductivity really doesn't matter when you are talking about differences of 0.XX degrees.
I've had a Nitro+ Vega 64 for 4 years now and have been dealing with this issue the whole time to varying degrees. I've used kryonaut for the entire time due to it being a very thick paste, but switched to Gelid GC Extreme just recently and it's been the same. Would always be the same story - repaste, then watch hotspot delta start to run away as the weeks go by and paste is "pumped" out from thermal cycles. (granted, it was never so bad on my card, just something I noticed). At stock settings temps were never an issue, but raising vega's anemic power limit also increased the hotspot delta disproportionately compared to edge temperature. Following the Vega specific hotspot/repasting guide from Igors Lab helped a ton - use thick paste, spread manually, followed by specific screw tightening order. But what helped even more was taking the cooler BACK OFF not long after repasting, observing how the paste had spread, and then re-applying the SAME paste (and potentially adding a little bit more if needed). It took me many attempts to finally figure out what works best, and temps have been great ever since. Along with deshrouding and mounting NF-A12x25's to the card, temps have been a complete non issue even running lots of voltage, though I run the card daily with a heavy undervolt. At 1.2v + 360w power limit on furmark, the card now has an edge/hotspot delta of only 20c, which is remarkable considering its 120w over stock. At the same settings but with a bad mount and stock fans at 100%, the card would eventually hit temp limit and shut down.
@@jerrywatson1958 and you show your "ignorance" for failing to read what he said properly. "Each tube of IC Diamond Thermal grease contains 7 carats of micronized diamond with diamond particle loadings @ 92% by weight." literally says it in the marketing slide and he's correct that you should not be using it for direct die cooling as any movement of the cooler and or through mounting pressure can damage the die causing micro-scratches in the die surface but too late for that now. now that we have that out of the way... as far as your original statement about it squeezing out excess kryonaut that's exactly what should happen. thermal paste is designed to fill the imperfections of the die/IHS and cooler contact surfaces, you want as much direct metal to metal contact and as little thermal paste as possible between it which is exactly what was happening in your case. but given your "ignorance" and blaming the thermal paste for doing exactly what it was suppose to do instead of likely your application process or mounting process leading you to use the improper thermal paste you wouldn't understand that. just saying... more than likely your issue was related to too much or not enough mounting pressure once the gpu had to carry the weight of the cooler after re-mounting it.
at 3:23 you see gray pad and blue/green pads. Ive had my merc 7900xtx for about a yr and iv never had a issue always just replacing the paste. This time i took off the blue\green pad and put gray pad in there place. I noticed instantly my hot spot went to 110c . Iv tried different thickness gray pads and same issue. Tomorrow ill get the same pads that i took off originally blue\green ones. I'm hoping this fixes this issue.
I had 107C hotspot and around 65-70C on normal temps on my Sapphire MBA card. I have since undervolted from 1150mV to 1120mV and locked core to 2400 min 2500 max +15 pwr and 2714 VRAM. With a custom fan curve I now have temps in the region of 55-62C and 70-85C hotspot. I didn't disasemble the cooler at all. I believe this is due to the phase change TIM used on these cards. I don't think AMD properly baked them in during testing. I am sure this TIM has a rather long cure time so people are buying these cards and seeing huge delta differences and getting scared. But I think after a few days or a week you will see these temps drop off and this delta get smaller and smaller.
If the manufacturer from the coolers use Honeywell PTM7950SP solvent based screen printing phase change paste then it needs at least 15 hours of curing time at room temperature before the heatsink gets attached. The process can be accelerated by using a forced air oven at 50'c for 4,5 hours. Devices equiped with Phase change material need some thermal cycles to achieve it's full potential.
@@realbeetlejuice I wouldn't panic at all, just make sure to run some tests and see if you model has a problem. But new drivers have been excellent and fixed a number of issues already. Repasting GPU is an option but not really needed, and undervolting seems to be the way to go. I would stress though synthetic benchmarks like 3DMark for me run and pass at 1065mV but a lot of games just outright crash after a few mins. So tweak your voltages.
@@Fezzy976 The "mistake" many people make is by replacing the factory paste with low viscosity (high silicone content) easy appliable paste like MX4 or NT-H1 because it's so good on all those 1st day temperatures on UA-cam reviews but what is not taken onto part is the holdup time. Modern age direct-die devices with high turbo speeds and the temperatures that come with it degrade many pastes very fast due to silicone separation or pump-out on uneven surfaces.
How did you apply the thermal paste - did you apply it in the middle in the form of small dots (the classic method), in an X shape as some people do, or did you spread it evenly across the entire surface with a spatula?
Just out of curiosity, how many watt power supply are you using in your test vs the original owner? I'm curious if you had a weaker power supply if the current would raise if the voltage sagged causing additional heat on the card. I caught on to the owners second card didnt have the same issue, but it still makes me curious.
Well also another thing is that you card is upright but in a system the card might be sagging and slightly changing the pressure contact... Maybe? Not very likely but that's something to be checked as it's a known issue with modern cards.
No, the thing is that if you pump more watts through it then the default wattage, THEN the delta between the GPU temp and the junction temp starts to grow to an abnormally high number. It's simply a matter of a cooler that is just average. You can easily test this with a game or synthetic test with much higher clocks then you did.
if you want more info.. my ref 6900 xt has the opposite orientation problem.. i have silverstone fortress 2 which have 90 degre rotate motherboard..so the gpu hangs straight down..vertical.. my hot spot was crap 110++.. then i remember certain vapor chambers 10 years ago had problem with orientation so i just tipped the case on its side(gpu sits horizontal) and hot spot temps are no longer a problem..
I was hitting 110c hotspot a lot before. was at 50c+ difference between hotspot and gpu. Undervolted the card to 1060mv and also tightened the screws. The screws were not fully turned tightened down when I took off the backplate. Hit a stable 2800 boost clock/2650mhz memclock after. I have a reference Asrock 7900xtx. Also updated the driver to the latest one. It does have huge delta sometimes on a few games while on others the delta were within 20c. Maybe the thermal pads and paste didn't set or something. Playing yesterday hitting 70c gpu ~ hotspot high 90's now.
I repasted the card with kryonaut and there was nothing improvement. Junction temeperature was 110 C before and after too but i have to mention my settings. 2600mhz, 1135 mv, +15% power consumption.
High hotspot isnt suprising. Had the exact same issue with my asrock 6900xt and had to drop voltage to 250 watts to keep temps under control. Finally repasted it with kryonaut extreme and problem was solved.
I have the reference 7900 XTX model shipped directly from AMD. I'm also experiencing a 30 degree delta with down clocking. However, I noticed the first thing you did was remove the backplate. Please consider testing the card again in a case with the backplate installed; which how most people would be using this card.
he did if you watched the entire video, it didn't change anything for him. either way i have a feeling this is a software or firmware issue causing a higher power draw some where in the die . if it was possible to see which exact sensor in the die was showing the 110C reading some one could figure it out but that'll likely never happen.
@@sirmonkey1985 I have the card in a case, not a open air test bench. This is not a blower style card, which means all of the heat escapes from the sides of the card into the case. Case fans can only help so much, especially if the side panel door is closed.
@@kucing1087 You are correct in this scenario because the test is on a open air test bench, not inside a case. The backplate can make a significant difference for cooling if used with thermal pads as a heat sink. This card does not do that. Mounted in a case, the backplate is more of a "heat trap" than a heat sink because of the design AMD chose.
I have the issue, repasted and its fine in stock settings, however if I run timespy with any added power it slowly ramps up to 40-50° delta. try adjusting power to max and rerun timespy, thats how I simulate it. I repasted with noctua nt H2
I have the same issue on an Asus rog strix 3070ti, i had to open it up, apply some new thermal pads and more thermal paste to fix the issue bechause Asus has bad quality assurance, so the pads wasn't aligned properly. On that card the issue was one of the VRMs being allmost without contact to the headspreader.
@@universalsinewave7559 No, that's just part of the construction of it, what you COULD try is packing thermal paste all over all the coils actually, the high frequency sound waves should be dampened by it!
kinda sad to see when the 5700xt came out there were huge thermal issues even mounting issues with cards like the STRIX/TUF models and the reference models had very high VRAM/VRM temps too. Which were co-opted through HP/OMEN and many 5700xt oem cards were sold through that venture as well....many im sure that were never undervolted/tuned to run well that are now at 2-3 years old on their last leg...barely running at 1900mhz.
In my setup, my MBA 7900XTX has no problem at stock at staying cool in very stressing games like e.g. Metro Exodus: EE. However the second I touch the power limit, the hotspot temp jumps up dramatically and maxing out fans doesn't seem to do much to help. In Exodus specifically, setting the power limit to +7% gets me in the mid 90s on junction and 10% makes it slowly rise up to 60c/110c (50 degree delta!). Bought some PTM7950/Gelid pads for the backplate to see if things can be improved. Maybe past ~370w the vapor chamber hits dryout (if that's something that can even happen)?
Have you tried mounting you GPU vertically to see if it helps. It was the only way I could leave the default power limit without thermal throttling hard.
Soft power play table might be messed up for the PL on the GPU, the power scaling going out of control when it is turned up may mean it's misconfigured in the bios? But yeah the reference heatsink might need to be examined, maybe insufficient liquid in it. It would be strange if a referenced liquid cooled version of the GPU still gets the issue.
I did run Port Royal in stress test window mode & power +15% and my 7900XTX does over 50C Delta in no time. Other comments seems to point the same. Replacing thermal paste on a new GPU can, maybe, fix this. I will request an replacement now that i know there is other cards that don't have this "issue". Clock speed dropping while increasing Vram frequency also is an concern.
My 6900 XT also has a delta of about 35-40° C at full load, I just thought this was normal as the card didn't seem to clock down, should I be concerned?
I have been getting the 110C junction temp issue in some games and after watching this I assumed I was going to have to RMA the card. However today I swapped out my 3m Display port cable for a 1.8m HDMI 2.1 cable and I now have junction temps in the 80c- 85c range.
With Radeon 7 I once had a bad mount or bad paste and got similar behaviour. Once I cleaned, re-pasted, and re-mounted my block it was better. Now mind you, a Radeon 7 always got 30C delta because of the weird HBM packaging but 40C+ was bad and could be addressed without too much effort.
Just opened mine today, re mounted, paste it came with is almost completely dry and poorly spread. Left as is to see if simply remounting would make a difference, it did not. Buying thermal paste tomorrow to re test and getting a vertical mount to try that after. I will update once done. Update: bought some Kryonaut thermal paste, re mounted and pasted thoroughly. It seems to have made a difference at first but after about an hour or so of heavy gaming, it peaked at 110c again. It seems to be a bit better since its no longer constantly at 110c, it oscilates anywhere between low 80s all the way up to 110. Pretty strange behavior. Any ideas for other things to try, if anything?
It was the power color reference model. It did not work, repasting was a very temporary "fix". Sent it back, got it replaced by another reference model and it came with the same issue. It is now in the process of being replaced by a Hellhound version. Hopefully this one doesn't have the same issue.
Let me add my 2 cents here I own Powercolor reference 7900 XTX (mounted vertically in Louqe RAW case) Out of the box it used to go up to 110C quite fast -- but it turned out the driver had the clock speed set to a whopping 3050 MHz Once I had fixed it back to the stock 2300-2500 MHz the junction temperature went down to 80C and stays there ever since. But in this video I see that the recorded footage shows regular 2400 MHz with junction still hitting 110C. Weird.
My card is the ASUS 7900xtx reference model and my case is NZXT H1 V2. In the normal H1 V2 orientation, where the GPU is mounted vertically (the three fans lined up vertically, like the AMD promotion picture for 7900 xtx), I can see the temperature delta as big as 50 celsius, junction temp reaching 110 celsius but GPU temp sitting at only 60-ish, fans running at 100%. Putting the case down to let the GPU sit horizontally (like on a test bench), everything is reasonable and the temperature delta is less than 20 degrees Celsius.
Temp delta should be less than 20C else do a thermal repasting/replacement of thermal pads. My repasted 3080ti temps repasted 2 years ago is creeping up but still ok. After repasting, temps were 65C/76C. Now it is 68C/83C.
@Der8auer 9:10 I have this card, reference from AMD Direct, it has this issue. It is currently still unopened/altered in any way from stock. I've posted some about it on my Twitter (seen in the media section) if you're interested in seeing results of it. The problem for me is easily reproducible -- UHD gaming in literally any slightly demanding title from Civilization VI to Minecraft to any First Person Shooter I have in my library causes 110-113c hotspot temps. I have initiated a RMA for the card, but they are yet to respond to the latest inquiry as of 2 days ago. (likely due to holiday hours) AMA you want/need I'd be happy to oblige (excluding opening it, I'm not comfortable with that, which is why I initiated an RMA
For those interested in the orientation, For my card, the "normal" orientation is 90°, "vertically" mounted, that being you'd turn the motherboard 90° with a normally vertically mounted GPU. I've tried normal orientation, I've tried Benchmark Orientation, the only change in temperature is when it's in a "normal" orientation (plugged directly into the motherboard), with the motherboard flat on a desk, like a Benchmark Rig may be placed. Doing so brings the temperatures under load from ~110°c to 102-106°c.
Yeah Halli, they did - it was about a 3 day waiting period, they offered either an RMA replacement or a full refund. I opted for the refund Wasn't all that impressed with 7900xtx's gains over a 6900xt About the same performance boost as I'd get from upgrading my CPU and platform to 13th gen intel from AM4 so I went for that instead
@@Justifier I take it you play in 1080p? I was thinking of getting one for my new 4k screen since my 3080 isnt keeping up all the way Btw do you think a cpu 3900x would bottleneck the 7900xtx in 4K?
@@ermirhalitaj5346 I no longer have my 7900xtx, as I opted for other hardware I played at both depending on the title Obviously, I **greatly** prefer 2160p, however if RayTracing is involved, or it's a newer title, I dropped it down to 1080p to get 120+ fps I don't think that a 7900xtx would lose much performance at 2160p paired with a 3900x, but likely would at 1080p with the exception of RT titles, where the lower ram speeds would hit performance significantly, but I could be wrong and anything pre-5000 series may You'd likely want to look about to see if you can find any similar pairings and come to your own conclusions on the specific titles you enjoy
i have a xfx 7900xt that has a hotspot of 90c and after changing the paste of thermal grizzly kryonaut changed nothing. still the same hotspot temp. I would like to drop the temp but unsure of how.
remember the inno3d ichill frostbite 4090 video ? well I got one and it had 61°C with 112°C hotspot >< dismantled it, 1/3 of the block above the gpu was mirror clean while the gpu had paste all over, repasted re-tightened with thermal grizzly kryonaut and now I have 46°C-56°C with 44°C memory...for 442w O_O amazingly cool card a bummer inno3d messed up the mounting (and maybe uses the wrong thermal paste brand ;) otherwise I would recommend 10/10 this monster card, alphacool made a killer block and the 4090 are uber efficient in some games my overall power went from 830w to 550w@psu for a much smoother gameplay, the best gaming experience I have ever seen since...forever
Is it really running hot, or is it a bad thermostat being put on these? The vapor chamber test would be to switch coolers from a known good GPU and slap that sucker on a hot GPU. If that fixes it then it's the vapor chambers for sure!
These are on-die temperature sensors, so there isn't anything to "put on" particularly that would cause the malfunction. Is it possible that the designed-in temperature diodes or sensing logic has some sort of a defect? I don't see how it would pass validation to begin with then, or how there isn't more damage, but i guess it would be silly to dismiss the possibility out of hand either. But seems exceptionally unlikely to me, more likely something with cooler subassembly.
I had issues with high overal temps with hotspot reaching 110C on my rx 6700xt pulse. I even bought morpheus cooler but when I wanted to install it ive realised the standoffs are too long so i packed it back and now the temps are much better with stock cooler. Apparently it was just badly mounted and reseating fixed it. Also ive first screwed cooler to card, gpu core bracket was last to screw in.
How do you test your card? Horizontal or vertically mounted inside the pc case? Because AMD says if you mount vertically you're fine, but if you mount horizontally as normal in most pc cases then your stock card maybe can overheats. I have no idea why vertical or horizontal matters but that's what AMD says.
I think that might be becuse of the vapor chamber, if there is a kind of over pressure thing and if because that heat transfer with pipes or die starts to goes away this behavior can be happen, and the score issue with ASRock card possibly because of VRAM frequency.
My Sapphire 7900XTX has no issues, temperatures are very similar to the values you got. It's still got the stock thermal paste. I have now applied the auto undervolt setting in the AMD Adrenalin drivers, but it hasn't made a great deal of difference
I just put liquid metal on mine, as if it were a PS5. I took all necessary precautions and now I get 60ºC edge and 70ºC hotspot while playing Horizon Zero Dawn and at maximum settings, of course.
should have tested his card before you changed his paste to verify the issue was still there as he said. also playing the same game as he showed may have helped reproduce the issue, as it may have been software related. perhaps some 3rd party app was incorrectly adjust the cards fan speed or disabling them all together?
I had the same issue with my card. It would reach 110C and I guess it must have went further because it happens two times where my PC completely shutdown to protect the card. I was running the 22.12.1 drivers. Then, they released 22.12.2 and everything seems to be back to normal now with deltas similar to what you have. However, something tells me that probably the thermal paste and pads need to be changed, maybe the pressure if not good enough; comparing my 3dmark results with people who have the same CPU, I found my results to be quite low in the area of 21500-22000 (OC) whereas some people have been getting 27000-29000. If I don't OC the card, the results are even lower, around 19000. Not sure what is going on.
One thing I want to add: the previous owner used his card vertically! Because a lot are drawing quick conclusions that "it's only because it was not used horizontally". It doesn't seem to be this simple :)
I organized two more completely stock cards today and will do a follow up with more testing soon - including different positions.
Could it be a problem with just MW2? There's atleast one other card online showing this problem in that game.
@Chris D I don't think one game changes the hardware installation...it's temperature not blue screen
@@royboysoyboy I hear you there but I'm thinking more along the lines of what happened to the 3090's and new world. Just hammering one part of the card in that case to failure, but here the card is keeping its self in "acceptable" limits.
You must test different positions; go look at all the Reddit posts - people report 110c in one position and then 80c in a different position, so the orientation matters
@@grim5116 will do that :)
Kelvin and Celsius is the same if you are measuring a delta between more than one temperature. They are linear so he is correct in saying Kelvin.
Okay, I was wondering about that, but it was still really bothering me each time he used it. >_
I am also German and we often use kelvin to emphatise that it is a delta value
I guess thats fine, but i have no reference to kelvin, since i don't use it. If he said 25celcius, i would know right away without thinking, but when he says 25kelvin, i have to check what it means.
Basic science isn’t taught in schools anymore 💀
@@skjetnis You learn something everyday
The problem was completely missed in the video.
I have tested my reference rx 7900 xtx and have observed that following:
Traditional orientation (horizontal with fans facing down) yields hotspot temperature of 110 C
Vertical mount (like shown in the video) the hotspot temps range between 80 C to 90 C
After testing and verifying these results, I performed an additional test where I took off the backplate to tighten all the mounting screws. Most of the screws were already sufficiently tight, but some had a tiny bit more turning capacity by hand. The results after hand tightening the screws were the same, with the horizontal mounting orientation yielding major temperature issues.
The mounting orientation yielding such drastically different results means that the problem is either with bad mounting pressure on some units or bad vapor chamber on some units.
I would love to see more thorough investigations as to why mounting orientation is causing this issue and what can be done to fix this issue, other than vertically mounting the GPU, which is not possible in most PC cases without modification.
Please consider these recommendations and I really hope to see another video update from you regarding this topic.
I tested my card exactly the way the previous owner used it :) he did not have horizontal mounting. But I will look into this. I don't think it'd this simple tho
@@der8auer-enI have this card, I had major hotspot issues when the card was mounted vertically (not traditional vertical though, I had mine with IO pointing to the floor). I then changed to traditional vertical mount and the temps were then ok.
Mine was vertically mounted in my meshlicious and had the issue. I did not bother to repaste tho. I just returned it. Laughable that so many cards have this issue from the factory.
@@der8auer-en Could it be possible some coating in vapor chamber got loose and it's moving around, blocking something and after taking card and moving it around it unblocked?
I have the reference 7900 XTX mounted horizontally. With stock paste and mount I was at about 15 to 20c delta, maxing in the low 80s for tjunction. I decided to repaste for fun with kryonaut (not extreme) and it maybe lowered both temps by 1 to 2c, so the stock paste and mount was actually perfectly fine.
For me, the issue only becomes visible once I unlock the powerlimit from 350 to 400W. Before that, the delta is quite big (30°C) with junction in the 90s but after that, it can go from 60-70°C edge temps to nearly 50°C delta with 110°C junction temps that cause throttling. This is a card in Hamburg, Germany. Could get that one to you - but I think it'd be better to find a card that shows a bigger delta out of the box.
as I explained in a comment I had the same as you with a pre-waterblocked 4090 factory I had 61-112 !!! gave my warranty up and opened it, a good 3rd of the gpu position on the block was clear, wasn't touching, cleaned repasted with thermal grizzly kyronaut and it's now 46-56°C for 440w I doubt I'm the only one who got a bad contact/paste videocard
@@fredEVOIX Just opening a GPU is not modification, and will not void your warranty alone. Granted, AIB's will always err on the side of caution and say something along the lines of, "retention bracket screws scratched." That is still not sufficient evidence enough to void a warranty under Moss Act, if residing in the US. Small claims court filing fee is usually few hundred dollars, small price to pay if you're replacing a $1600 card.
another Amlie no trace user
@Jonas Jonaitis Except, it's economically not small whatsoever, which is the context here.
Europe has even better laws to protect the consumer.
So what is your point?
@Jonas Jonaitis It's the same in EU, dude.
Try to mount it horizontally like in a normal case, thats when most people get issues with hotspots. Vertical mounting gets rid of it from what I've heard.
Reddit has discussion threads where changing vapor chamber orientation changes hot spot behavior.
This. Someone on reddit claims the temps are normal if he lays his PC case on its side, but if his case is upright the problem returns. I imagine der8auer is using a test bench instead of a normal PC chasis, which might explain why he didn't have the issue his friend did with the card.
This is absolutely true.
I have tested and verified horizontal vs vertical mounting orientation and observed that the problem goes away when mounted vertically.
@@nerdstrangler4804i can just add that the original owner used it vertical. Obviously I did the same. So it's not limited to horizontally only
@@der8auer-en Remember that screwing down the card's metal bracket to a case can change the tension distribution across pcb and cooler. You have to take this into account in further analyses.
Would be interesting to see if the orientation of the card would change the result. Vertical mount (as tested here) vs. horizontal mount (as it would usually be mounted in a case).
I agree, it could be a horizontal bending issue. And there could be a high temperature in the user's case. I didn't see the GPU temperature from the user when the hot spot comes to 109 degrees.
If 7900xtx has the anti-gravity heat pipes then the difference should be a few degrees C, if not probably >20 C. But it shouldn't change the delta too much between GPU vs GPU hotspot
Will look into that :)
I had rear some reports about temperature problems when mounted horizontally that went away with vertical mount, so if roman uses only testbench he wont get same results as if it was in case.
@@JokerjOkER-du8br the problem is when a card is sagging in horizontal orientation without support parts of the GPU may lose contact.
My reference card had 110° hotspot and 60° gpu temp. 3000rpm (100%) fan and downclocking. I just returned it. I'm perfectly fine with changing thermal paste but I'm not doing it with a brand new card of this cost. Completely ridiculous.
I overclocked my 4080 +150 on the core and +600 on the memory (founder's edition) The GPU temp doesn't reach more than low 60's and the memory hasn't hit 70c yet.
@MultiNastyNate exactly. Have a 4080 now to. Fans don't even turn on playing warzone 2 at 1440p 200+fps lol. Sits at 65° 1300rpm fan in games that max usage.
There was some talk on reddit about the possible issue being people using a non certified display port cable causing the issue, related to the 20pin in the cable. This could be one of the variables as to why Der8auer could not replicate the issue
The orientation of GPU is important as well, if its a test bed, then most probably you wont see a problem, people reported that tilting the case to its side like a test bed results in lower junction temp, so if you are doing it on a test bed, then just try installing the card horizontally (fans facing down) and the case sitting vertically like a normal PC and then retest on new affected card since the one you have already got fixed by re-applying thermal paste and balancing the pressure.
I was also thinking about whether it was in a case or on a test bench, which is usually -- if not always -- open-air. That, combined with the orientation, will definitely impact the air flow.
I think i read something about it on Reddit or TPU. You might be right.
I personally owned a single fan gigabyte gtx 970, it was running 60C in the mobo and 88C in the vertical GPU bracket, that was when i found out about this issue, it was easily reproducable by tilting the case on its back or face. Vertical bad temp horizontal good temp
I’ve got data on this. Orientation creates a higher temp delta. It’s crazy.
Test bench style mounting is okay
Normal case mounting causes 110 Celsius junction temp.
You people keep saying orientation matters but Roman already commented in this comment section saying he has tried both orientations and it is irrelevant.
Yes, that was one of the theories was that mounting pressure was probably the most common cause as they were rushing to get enough stock out for launch. But it is still great to have this confirmed by someone like yourself. Thank you. (I have one and no issues what so ever. Even OC and u-v very well).
Well, mine‘s a bag of sh!t
Remind me of loot box
@@acidgaming0 sorry to hear that. I wish we had a rough percentage of product with issues. I guess Powercolor is collecting this data for AMD. Very sad indeed. I hope this gets resolved and you get looked after asap.
@@ktpt3732 It does sound that way. Some of us very happy with their product, others, like acidgaming0 above, received a bad model. Very sad. I hope this gets resolved and these people get looked after asap.
@@grizzleebair Thanks, I hope AMD will look into this after the holidays.
My XFX 7900XTX reference (horizontal) would reach 110C hotspot within 6 minutes~ of running furmark. I did open it up to repaste with MX4 and did torque the screws as far as they would go by hand but unfortunately my temperature issue persisted. I did test vertical mount by flipping my case over and it did definitely have dramatically better results. At this point I don't know what to do besides a RMA since I plan to use the gpu horizontally.
Also when repasting I noticed the stock thermal paste was very solid, best way I would describe the consistency is very old and worn out thermal paste.
It's almost as if the card is designed to reach 110c.
@@2buckgeo843cool
@@2buckgeo843 Because your 4080 is a premium product.
If you play a game, it probably won't be that hot, it's because you are stress testing it.
My 5700xt strix displayed similar temps, turned out the thermal pad on the fets was a whole 1mm to thin so the screw heads made no contact to the back of the pcb, simply folded the pad and it dropped junction temps from around 108 to high 80’s. Also the die was pretty scratched from the cooler sitting crooked for so long and possible having a loose cooler when in shipping.
Shiiit you'd think ASUS would have better QC, but that's how things be now huh.
@@SianaGearz I was told by the ASUS rep it was AMD’s fault for sending out faulty measurement’s and I could just send it in but I wanted to see if I could fix it first or figure out what they did wrong but yeah so much for q&a the screw heads were literally a whole 1mm from the back of the pcb on VRM side.. lol
I could guess the issue might come from the squishy pads on the memory chips that could possibly affect the pressure on the GPU die, now when they been opened and messed around with they probably been flattened more and the pressure on the die probably increased again
Not sure why thermal putty isn’t used as it drastically better…
@@christophervanzetta if the pads on the AMD cards would be the problem it could be so easy as to just apply pressure around this area by hand to push the card down to a flush level. Unless the paste get dried up from the high temperature very quickly
Thanks for your efforts on making this video. I have been experiencing those issues with my recently bought 7900xtx sapphire reference mode. At first i thought, that i was installing in the wrong way my driver or my case's fans weren't working properly. I hope with videos like this, even that is going to cost a lot, AMD should accept RMA's and learn from this experience. And don't cheap out on the thermal paste in a 1k USD/EUR GPU.
This is unfortunate because the cooling solution including the paste used on the die is superior to what they used in the previous gen (check GamersNexus's teardown review); however, you can have all sorts of problems in the assembly line. These should be caught in QA test but obviously, they weren't.
So i had exact same problem with my 6700XT. Weirdly enough problem was solved not by thermal paste but by adding support bracket to a card. It somehow bent itself to the point that part of GPU lost contact with a cooler i guess.
what exact card do you have? I recently bought MSI MECH 6700XT and same issue here. Just asking because maybe I can fix mine too without sending it for RMA.
@@jachupol5823 heatsink sag, just add a "GPU support stick" (tm)Galax
@@jachupol5823 I had aorus from gigabyte. You can put a book under the card as a temporary support and see if it solves the probem.
@@PanaehaliTut alright I will try that lol
@@PanaehaliTut also thankfully it's not as bad for me, the core to hotspot delta is only like 30-35C so it's only dropping clocks slitghtly
When you RMA a GPU and they test it, they will most likely position the GPU vertically as the testing board they use it on will be flat on a table... This also means they don't notice any problems that you would see if the GPU is installed horizontally in a case.
I’ve just put my Powercolor 7900 XTC MBA in my Pc and ran kombustor. After 1 minute my junction temp hits 110 Celsius and the GPU is at 59 Celsius. Even with fans at 100% it takes 2 minutes to get to the 110 Celsius junction temperature!
On the AMD forums (I can't provide actual links because UA-cam deletes my posts if I do), one person tested his card in four orientations (look for "7900xtx reference orientation testing. Orientation impacts performance"). The bad temperatures show up when the card is a "normal" orientation (i.e., when the card is horizontal). In a "test-bed" orientation (when the card is vertical as with your setup), the temperatures are fine. Try tipping your test-bed on its side or putting the card in a regular, tower-type case.
I was re-pasting my 2 y old 6900xt yesterday because hot spot was reaching 105* C but GPU temp was never above 64* C
after doing this the hot spot temp stops at 82* C but the GPU temp goes up to 71* C
I was surprised because the original paste looked good and was still quite soft
The reason for the higher GPU temp might be that the heat is more spread out now, rather than sitting just within that hotspot.
@@DigitalJedi Nah it's because the card before repaste hit the tjmax at hotspot so it would clock down. Higher temperature in this case means more performance due to higher amount of workload being performed.
High hotspot temp could mean the chip is not fully covered in thermal paste. Bad news if the die has no heat spreader. Heard of this happening on nvidia gpus as well and that it affects some amd 6000 series as well.
Hello could you tell me what is the proper mounting pressure for a 6950xt reference card?
I encountered this issue with my reference 6800XT when it was vertically mounted with ports up - hotspot temps would be through the roof and half the radiator would be cool to touch.
Thanks for this video. I just stress tested my XTX. Results here as a point of data for others:
65 degrees F in room
GPU temp never rose above 65 C and junction temp which has been the hotspot on some of these cards never rose above 79 C even under sustained 100% load.
The fans never went above 62% to maintain those numbers.
How long did you test it? I have your numbers as well on my card
@@davideresentini6826 I ran it for 15 minutes.
I have reference XTX bought from AMD's site in US. My junction temp rarely reaches 90°. It seems that the hotspot problem is more common with models sold in EU. Anecdotal evidence only, but I've certainly seen more EU users complain about the hotspot temp problem.
derbauer fixed the card just by looking at it. Incredible.
This guy is an absolute legend.
air pocket? I've noticed it near impossible to see signs of one after you pull the heatsink as it suctions away any traces, unless the thermal paste is old enough to leave a bubble mark.
I always make an 'X' with a little extra in the middle to ensure the air gets pushed out. With Liquid Metal i put an even coat just enough to cover the surfaces with surface tension then a little extra on the middle of the GPU/CPU and the middle of the heatsink. Over time the air pockets might pop and temps will go to normal, but that can take weeks of temperature fluctuation to work itself out
Is GPU sag from horizontal mounting vs vertical mounting a factor? Also were the screws possibly not torqued enough? It seems like either disassembling and reassembling the card or mounting it vertically corrected the problem.
I haven't heard of it being a problem, those screws are made to be torqued until they hit a stop and not force further, the torque is given by the metal bracket (spring). The card can't be easily influenced by sagging as it's got many screws which keep it from flexing, especially with the help of the backplate. Most of the visible sagging would come from the PCIE slot itself plus a little bit from the edge of the PCB which inserts into the PCIE slot, but that little bending there is just local and doesn't reach the GPU area. In this case, I would just assume the owner either mounted it incorrectly or used a bad TIM or didn't clean the chip well enough before reapplying TIM. Also, the issue might not manifest with normal PL in place, might need to give it maximum power to see bigger Deltas.
9:32 maybe it only happens in certain games like Call of Duty MW 2. That game runs abnormally well on these GPUs. Like beating a 4090. Maybe it's using silicon that other games aren't using. Maybe Furmark is not activating whatever is causing this.
how so? this is a gpu not a cpu, its not like you have lots of circuits to chose from.
@@humorss Well there are, there's the shader cores with separate ALUs for different kinds of arithmetic operations, and of course even in one multipurpose ALU, all operations don't hit the same exact semiconductors, and around them there's TMU (texture fetch), memory/scheduling (because memory fetch often forces latency hiding logic to swap tasks), and ROP. And that not even counting the new RT hardware. When you read utilisation, it just takes the max of all values per SM, and any operation in an ALU is considered as that ALU being utilised, in spite of some being potentially more power hungry than others. Typically, ROPs are severely underutilised in most of the render passes.
I cant wait to see the mods we can do to the reference cards!
My reference card is also getting 110 Celsius hotspot temperature. I've tried already new thermal paste, used different bracket pressure on GPU by adjusting GPU brackets crews. No difference so far. I'm waiting what you will discover. I'm curious what is your method of placing bracket on the card. Which pressure did you used. Thanks for looking into that problem. Can't wait for that video.
Try using graphene kryosheet instead of thermal paste
@@rodiculous9464 or ptm7950, works wonders for gpus
Hello. I also have a 7900XTX XFX Merc 310 card, the hot spot cannot even be cooled with a water cooler. The temperature of the hot spot was still around 110 degrees and I did what I could (various thermally conductive pastes, tapes, etc.). I'm quite stressed about it, because the card doesn't cost a lot of money. I would also like to point out that there are of course different temperatures in full HD, 2K and 4K at certain Hz, where 60Hz will have low temperatures. Where, of course, the high temperatures are resolved at 144 Hz in 4K! That's pretty important!
One thing that many people think is that in a vertical mount in a test bench it’s fine but mounting it horizontally in a pc the cooler pulls away from the pcb and die slightly. I need to finish the video before I comment i see you talked about this issue above but i will watch it all love your content.
Thanks for this, I had already repasted my RX 7900 XTX and it did help a little (96c junction temps at stock) but I just did so again with kryonaut extreme and also made sure tigten everything down correctly so far at stock settings max junction temp is 86c, biggest delta I've seen is 21c!!!! Yay!!!!!
How long was the card stressed? If they're using phase change material like the honeywell pads, temps will be higher than normal for around the first 24 hours of use.
The problem is that the GPU was mounted vertically on a test bench, most users only experience issues in the horizontal position.
The previous owner did not have horizontal mount. I will Test it but I'd not draw a conclusion from this right away
@@der8auer-en hi, I’m retesting after running stock settings on my card and in time spy extreme are hitting about 90 degrees on junction and 61 on the gpu temps just wondering if this sounds about right, closed case about 25 degrees room temps
Try to run card horizontally mounted. Seems that some cards has some vapor chamber issues and cards works better when verically mounted (as they are in test bench etc...).
Here's an update, I have a 7900XTX originally from Sapphire, but it's mounted with a Bykski waterblock.
The GPU works at around 45 degrees, while the GPU Hotspot reaches 110 degrees.
I'm already using the GPU vertically and I don't know what the problem is.
Is anyone else in the same situation?
Honestly that's the best case scenario. If anything AMD can send out some free thermal paste and post a video about how to disassemble the card and properly re-paste it. If people don't want to wait they can say you're welcome to re-apply paste on your own and we won't void the warranty. Not nearly as bad as the difficult to clip in 12/16 pin connectors.
Some people bought new PSU where 12 pin to two 8 pin connectors are given to them... i wonder if the 12 pin slots on the PSU outputs different wattage than normal 8 pin slots.
This card has some serious issues. Mine was running hot at 110 junc with a bad dp cable. Swapped cable - temps went down 20°. Tore the PC apart few days later. Again - 110°.. Guess what? Different temps depending on which displayport port you connect to. I can imagine that you didn't fix anything but rather don't encounter the problem because of some fluke. Interaction with display or whatsnot
Maybe it's the order you put the screws in. Maybe a certain order it bends the PCB.
i'll be interested in a video where you CAN reproduce the issue.. This one didn't quite have the meat on the bones i was hoping for
could there be a difference between your mounting pressure and his
my powercolor card will run at 397w for 2-3 mins, then the heat starts to overwhelm the card, fans go up to 2900 rpm, sounds like a vacuum cleaner ALL DAY LONG... wattage will lower to 330w. Horizontal mount is about 10° hotter than vertical mount, hotspots temps are 110° no matter what. Boxing it up and taking it back.
I have a reference asus 7900xtx and have had 40*c delta's in cyberpunk. I tried reproducing the issue in furmark but only saw a max temp of 90*C in furmark. In cyberpunk however after about 20-30 minuets of gaming my card had a 40*c delta and junction temp hit 110*'c with gpu temp at 70*C, when junction temp hit this temperature the card went from a 2500mhz clockspeed to a 2300mhz clockspeed and fans hit 3000 rpm. my card has standard horizontal mounting.
Thanks for this. I have a Sapphire Toxic 6900XT I bought this past July. It ran great OTB gpu temps in the 60's while gaming in the Division 2, hotspot temps in the low 80's. But within 3wks temps went up 10C on the gpu and hotspot was in the high 90's. After a week of temps hitting 100C, I took it apart and repaste the gpu with Kryonaut, temps dropped back to what I saw in the beginning. But yet again 2-3 wks later temps rose back to high 90's on the hotspot. I repasted again but with IC Diamond. It had appeared to me that the Kryonaut had been squeezed out. IC Diamond is a thick TIM. This fixed the issue for the last 3 months. I didn't have this issue with my cpu and Kryonaut. I guess it's just luck of the draw.
IC Diamond shouldn't really be used for GPU's or CPU direct die. As GPU's don't have a heatspreader and you are putting the paste directly onto the die it self. IC Diamond is just that, it has diamond particles inside it meaning it can scratch the die surface and MAYBE damage it over time.
@@Fezzy976 You show your ignorance with that statement. Maybe you should actually look it up on their website before you make ignorant statements of Bull. Nobody beats IC Diamond for thermal conductivity and it lasts for years. It doesn't get squeezed out by thermal expansion and contraction like thinner pastes.
@@jerrywatson1958 How am I being ignorant? I was just stating a possible outcome based on facts. This is the same kind of reason for not using Arctic Silver 5 or other metal based TIMs on direct die due to them being electrically conductive. Sure it probably won't fry your harware if you do it right but there is always a chance of human error. There is so little difference between most of the high end thermal pastes that thermal conductivity really doesn't matter when you are talking about differences of 0.XX degrees.
I've had a Nitro+ Vega 64 for 4 years now and have been dealing with this issue the whole time to varying degrees.
I've used kryonaut for the entire time due to it being a very thick paste, but switched to Gelid GC Extreme just recently and it's been the same. Would always be the same story - repaste, then watch hotspot delta start to run away as the weeks go by and paste is "pumped" out from thermal cycles. (granted, it was never so bad on my card, just something I noticed). At stock settings temps were never an issue, but raising vega's anemic power limit also increased the hotspot delta disproportionately compared to edge temperature.
Following the Vega specific hotspot/repasting guide from Igors Lab helped a ton - use thick paste, spread manually, followed by specific screw tightening order. But what helped even more was taking the cooler BACK OFF not long after repasting, observing how the paste had spread, and then re-applying the SAME paste (and potentially adding a little bit more if needed). It took me many attempts to finally figure out what works best, and temps have been great ever since.
Along with deshrouding and mounting NF-A12x25's to the card, temps have been a complete non issue even running lots of voltage, though I run the card daily with a heavy undervolt. At 1.2v + 360w power limit on furmark, the card now has an edge/hotspot delta of only 20c, which is remarkable considering its 120w over stock. At the same settings but with a bad mount and stock fans at 100%, the card would eventually hit temp limit and shut down.
@@jerrywatson1958 and you show your "ignorance" for failing to read what he said properly. "Each tube of IC Diamond Thermal grease contains 7 carats of micronized diamond with diamond particle loadings @ 92% by weight." literally says it in the marketing slide and he's correct that you should not be using it for direct die cooling as any movement of the cooler and or through mounting pressure can damage the die causing micro-scratches in the die surface but too late for that now.
now that we have that out of the way... as far as your original statement about it squeezing out excess kryonaut that's exactly what should happen. thermal paste is designed to fill the imperfections of the die/IHS and cooler contact surfaces, you want as much direct metal to metal contact and as little thermal paste as possible between it which is exactly what was happening in your case. but given your "ignorance" and blaming the thermal paste for doing exactly what it was suppose to do instead of likely your application process or mounting process leading you to use the improper thermal paste you wouldn't understand that. just saying... more than likely your issue was related to too much or not enough mounting pressure once the gpu had to carry the weight of the cooler after re-mounting it.
I still do not know if ALL OEMs have the same problem with AMD reference design or is it just some of them?
at 3:23 you see gray pad and blue/green pads. Ive had my merc 7900xtx for about a yr and iv never had a issue always just replacing the paste. This time i took off the blue\green pad and put gray pad in there place. I noticed instantly my hot spot went to 110c . Iv tried different thickness gray pads and same issue. Tomorrow ill get the same pads that i took off originally blue\green ones. I'm hoping this fixes this issue.
I had 107C hotspot and around 65-70C on normal temps on my Sapphire MBA card.
I have since undervolted from 1150mV to 1120mV and locked core to 2400 min 2500 max +15 pwr and 2714 VRAM.
With a custom fan curve I now have temps in the region of 55-62C and 70-85C hotspot. I didn't disasemble the cooler at all.
I believe this is due to the phase change TIM used on these cards. I don't think AMD properly baked them in during testing. I am sure this TIM has a rather long cure time so people are buying these cards and seeing huge delta differences and getting scared. But I think after a few days or a week you will see these temps drop off and this delta get smaller and smaller.
so I should wait to receive my sapphire ref 7900xtx and not panic?
If the manufacturer from the coolers use Honeywell PTM7950SP solvent based screen printing phase change paste then it needs at least 15 hours of curing time at room temperature before the heatsink gets attached. The process can be accelerated by using a forced air oven at 50'c for 4,5 hours.
Devices equiped with Phase change material need some thermal cycles to achieve it's full potential.
@@realbeetlejuice I wouldn't panic at all, just make sure to run some tests and see if you model has a problem. But new drivers have been excellent and fixed a number of issues already. Repasting GPU is an option but not really needed, and undervolting seems to be the way to go. I would stress though synthetic benchmarks like 3DMark for me run and pass at 1065mV but a lot of games just outright crash after a few mins. So tweak your voltages.
@@phenos Very interesting, thank you for the info. I thought it had some cure time involved.
@@Fezzy976 The "mistake" many people make is by replacing the factory paste with low viscosity (high silicone content) easy appliable paste like MX4 or NT-H1 because it's so good on all those 1st day temperatures on UA-cam reviews but what is not taken onto part is the holdup time. Modern age direct-die devices with high turbo speeds and the temperatures that come with it degrade many pastes very fast due to silicone separation or pump-out on uneven surfaces.
How did you apply the thermal paste - did you apply it in the middle in the form of small dots (the classic method), in an X shape as some people do, or did you spread it evenly across the entire surface with a spatula?
Just out of curiosity, how many watt power supply are you using in your test vs the original owner? I'm curious if you had a weaker power supply if the current would raise if the voltage sagged causing additional heat on the card. I caught on to the owners second card didnt have the same issue, but it still makes me curious.
Wow no issues after you touched it. :D
Well also another thing is that you card is upright but in a system the card might be sagging and slightly changing the pressure contact... Maybe?
Not very likely but that's something to be checked as it's a known issue with modern cards.
i had the exact same problem with an 6900xt AsRock. Changed the thermal paste and the delta went from 40 to 20.
65C and 110c on mine. completely stock. Beast performance too. Doesn't make much sense.
My card also goes to 110°C... Phantom Gaming XTX. Think simply remounting everything would fix it?
No, the thing is that if you pump more watts through it then the default wattage, THEN the delta between the GPU temp and the junction temp starts to grow to an abnormally high number. It's simply a matter of a cooler that is just average. You can easily test this with a game or synthetic test with much higher clocks then you did.
I have a gigabyte reference 7900XTX mounted vertically, my deltas were very similar to yours.
if you want more info.. my ref 6900 xt has the opposite orientation problem.. i have silverstone fortress 2 which have 90 degre rotate motherboard..so the gpu hangs straight down..vertical.. my hot spot was crap 110++.. then i remember certain vapor chambers 10 years ago had problem with orientation so i just tipped the case on its side(gpu sits horizontal) and hot spot temps are no longer a problem..
I was hitting 110c hotspot a lot before. was at 50c+ difference between hotspot and gpu. Undervolted the card to 1060mv and also tightened the screws. The screws were not fully turned tightened down when I took off the backplate. Hit a stable 2800 boost clock/2650mhz memclock after. I have a reference Asrock 7900xtx. Also updated the driver to the latest one. It does have huge delta sometimes on a few games while on others the delta were within 20c. Maybe the thermal pads and paste didn't set or something. Playing yesterday hitting 70c gpu ~ hotspot high 90's now.
The question I would like answered is what refresh rate the monitors were using as just about every reviewer is making a mess out of this.
I repasted the card with kryonaut and there was nothing improvement. Junction temeperature was 110 C before and after too but i have to mention my settings. 2600mhz, 1135 mv, +15% power consumption.
High hotspot isnt suprising. Had the exact same issue with my asrock 6900xt and had to drop voltage to 250 watts to keep temps under control. Finally repasted it with kryonaut extreme and problem was solved.
I have the reference 7900 XTX model shipped directly from AMD. I'm also experiencing a 30 degree delta with down clocking. However, I noticed the first thing you did was remove the backplate. Please consider testing the card again in a case with the backplate installed; which how most people would be using this card.
he did if you watched the entire video, it didn't change anything for him. either way i have a feeling this is a software or firmware issue causing a higher power draw some where in the die . if it was possible to see which exact sensor in the die was showing the 110C reading some one could figure it out but that'll likely never happen.
@@sirmonkey1985 I have the card in a case, not a open air test bench. This is not a blower style card, which means all of the heat escapes from the sides of the card into the case. Case fans can only help so much, especially if the side panel door is closed.
@@kucing1087 You are correct in this scenario because the test is on a open air test bench, not inside a case. The backplate can make a significant difference for cooling if used with thermal pads as a heat sink. This card does not do that. Mounted in a case, the backplate is more of a "heat trap" than a heat sink because of the design AMD chose.
So is 95-98 degrees on average safe then? Just got a gigabyte 7900xtx im scared of these temps getting so high
Io-tech just tested and they got 7 - 9c drop after installing card vertically. New paste and remounting didn't do much for them.
I have the issue, repasted and its fine in stock settings, however if I run timespy with any added power it slowly ramps up to 40-50° delta. try adjusting power to max and rerun timespy, thats how I simulate it.
I repasted with noctua nt H2
I have the same issue on an Asus rog strix 3070ti, i had to open it up, apply some new thermal pads and more thermal paste to fix the issue bechause Asus has bad quality assurance, so the pads wasn't aligned properly.
On that card the issue was one of the VRMs being allmost without contact to the headspreader.
My ASUS ROG 3070 Ti has coil whine issues. Idk if getting a new PSU will fix that issue.
@@universalsinewave7559 No, that's just part of the construction of it, what you COULD try is packing thermal paste all over all the coils actually, the high frequency sound waves should be dampened by it!
kinda sad to see when the 5700xt came out there were huge thermal issues even mounting issues with cards like the STRIX/TUF models and the reference models had very high VRAM/VRM temps too. Which were co-opted through HP/OMEN and many 5700xt oem cards were sold through that venture as well....many im sure that were never undervolted/tuned to run well that are now at 2-3 years old on their last leg...barely running at 1900mhz.
In my setup, my MBA 7900XTX has no problem at stock at staying cool in very stressing games like e.g. Metro Exodus: EE. However the second I touch the power limit, the hotspot temp jumps up dramatically and maxing out fans doesn't seem to do much to help. In Exodus specifically, setting the power limit to +7% gets me in the mid 90s on junction and 10% makes it slowly rise up to 60c/110c (50 degree delta!). Bought some PTM7950/Gelid pads for the backplate to see if things can be improved. Maybe past ~370w the vapor chamber hits dryout (if that's something that can even happen)?
Have you tried mounting you GPU vertically to see if it helps.
It was the only way I could leave the default power limit without thermal throttling hard.
Soft power play table might be messed up for the PL on the GPU, the power scaling going out of control when it is turned up may mean it's misconfigured in the bios?
But yeah the reference heatsink might need to be examined, maybe insufficient liquid in it.
It would be strange if a referenced liquid cooled version of the GPU still gets the issue.
Maybe people have bad cases? No or poor airflow? Horizontal mounting close to the glass? It's weird that it is such a big difference now!
I did run Port Royal in stress test window mode & power +15% and my 7900XTX does over 50C Delta in no time. Other comments seems to point the same. Replacing thermal paste on a new GPU can, maybe, fix this. I will request an replacement now that i know there is other cards that don't have this "issue". Clock speed dropping while increasing Vram frequency also is an concern.
My 6900 XT also has a delta of about 35-40° C at full load, I just thought this was normal as the card didn't seem to clock down, should I be concerned?
I have been getting the 110C junction temp issue in some games and after watching this I assumed I was going to have to RMA the card. However today I swapped out my 3m Display port cable for a 1.8m HDMI 2.1 cable and I now have junction temps in the 80c- 85c range.
With Radeon 7 I once had a bad mount or bad paste and got similar behaviour. Once I cleaned, re-pasted, and re-mounted my block it was better. Now mind you, a Radeon 7 always got 30C delta because of the weird HBM packaging but 40C+ was bad and could be addressed without too much effort.
Just opened mine today, re mounted, paste it came with is almost completely dry and poorly spread. Left as is to see if simply remounting would make a difference, it did not. Buying thermal paste tomorrow to re test and getting a vertical mount to try that after. I will update once done.
Update: bought some Kryonaut thermal paste, re mounted and pasted thoroughly. It seems to have made a difference at first but after about an hour or so of heavy gaming, it peaked at 110c again. It seems to be a bit better since its no longer constantly at 110c, it oscilates anywhere between low 80s all the way up to 110. Pretty strange behavior. Any ideas for other things to try, if anything?
Just out of curiousity.
What model do you have ?
Do you somehow managed to lower temps ?
It was the power color reference model. It did not work, repasting was a very temporary "fix". Sent it back, got it replaced by another reference model and it came with the same issue. It is now in the process of being replaced by a Hellhound version. Hopefully this one doesn't have the same issue.
Always good work on your videos
Let me add my 2 cents here
I own Powercolor reference 7900 XTX (mounted vertically in Louqe RAW case)
Out of the box it used to go up to 110C quite fast -- but it turned out the driver had the clock speed set to a whopping 3050 MHz
Once I had fixed it back to the stock 2300-2500 MHz the junction temperature went down to 80C and stays there ever since.
But in this video I see that the recorded footage shows regular 2400 MHz with junction still hitting 110C. Weird.
My card is the ASUS 7900xtx reference model and my case is NZXT H1 V2. In the normal H1 V2 orientation, where the GPU is mounted vertically (the three fans lined up vertically, like the AMD promotion picture for 7900 xtx), I can see the temperature delta as big as 50 celsius, junction temp reaching 110 celsius but GPU temp sitting at only 60-ish, fans running at 100%. Putting the case down to let the GPU sit horizontally (like on a test bench), everything is reasonable and the temperature delta is less than 20 degrees Celsius.
Undervolting should also bring the temps down and clocks up. The amd cards run way too hot for my liking with settings right out of the box.
Is it possible to do tutorials on how to disassemble and reassemble and the other things you mentioned to do for the card? Hope it's not a hassle.
The hotspot on my red devil is hitting 110. What can I do to get that down?
so only that specific one has this issue. and the issue can not be reproduced?
Temp delta should be less than 20C else do a thermal repasting/replacement of thermal pads. My repasted 3080ti temps repasted 2 years ago is creeping up but still ok. After repasting, temps were 65C/76C. Now it is 68C/83C.
@Der8auer
9:10 I have this card, reference from AMD Direct, it has this issue. It is currently still unopened/altered in any way from stock.
I've posted some about it on my Twitter (seen in the media section) if you're interested in seeing results of it.
The problem for me is easily reproducible -- UHD gaming in literally any slightly demanding title from Civilization VI to Minecraft to any First Person Shooter I have in my library causes 110-113c hotspot temps.
I have initiated a RMA for the card, but they are yet to respond to the latest inquiry as of 2 days ago. (likely due to holiday hours)
AMA you want/need I'd be happy to oblige
(excluding opening it, I'm not comfortable with that, which is why I initiated an RMA
For those interested in the orientation, For my card, the "normal" orientation is 90°, "vertically" mounted, that being you'd turn the motherboard 90° with a normally vertically mounted GPU.
I've tried normal orientation, I've tried Benchmark Orientation, the only change in temperature is when it's in a "normal" orientation (plugged directly into the motherboard), with the motherboard flat on a desk, like a Benchmark Rig may be placed.
Doing so brings the temperatures under load from ~110°c to 102-106°c.
@@Justifier Did they ever respond?
Yeah Halli, they did - it was about a 3 day waiting period, they offered either an RMA replacement or a full refund. I opted for the refund
Wasn't all that impressed with 7900xtx's gains over a 6900xt
About the same performance boost as I'd get from upgrading my CPU and platform to 13th gen intel from AM4 so I went for that instead
@@Justifier I take it you play in 1080p? I was thinking of getting one for my new 4k screen since my 3080 isnt keeping up all the way
Btw do you think a cpu 3900x would bottleneck the 7900xtx in 4K?
@@ermirhalitaj5346
I no longer have my 7900xtx, as I opted for other hardware
I played at both depending on the title
Obviously, I **greatly** prefer 2160p, however if RayTracing is involved, or it's a newer title, I dropped it down to 1080p to get 120+ fps
I don't think that a 7900xtx would lose much performance at 2160p paired with a 3900x, but likely would at 1080p with the exception of RT titles, where the lower ram speeds would hit performance significantly, but I could be wrong and anything pre-5000 series may
You'd likely want to look about to see if you can find any similar pairings and come to your own conclusions on the specific titles you enjoy
First, loving every upload you bring us... always informative regardless of what everyone else says.
driver feature. calculates more accurate number for some reason?
did you get the latest drivers, they were supposed to help a lot with this issue
Do you still have this problem theory? Or is there some other problem causing this?
i have a xfx 7900xt that has a hotspot of 90c and after changing the paste of thermal grizzly kryonaut changed nothing. still the same hotspot temp. I would like to drop the temp but unsure of how.
remember the inno3d ichill frostbite 4090 video ? well I got one and it had 61°C with 112°C hotspot >< dismantled it, 1/3 of the block above the gpu was mirror clean while the gpu had paste all over, repasted re-tightened with thermal grizzly kryonaut and now I have 46°C-56°C with 44°C memory...for 442w O_O amazingly cool card a bummer inno3d messed up the mounting (and maybe uses the wrong thermal paste brand ;) otherwise I would recommend 10/10 this monster card, alphacool made a killer block and the 4090 are uber efficient in some games my overall power went from 830w to 550w@psu for a much smoother gameplay, the best gaming experience I have ever seen since...forever
Could a solution be to just loosen all the mounting screws and spring plate and then retighten them starting with the GPU?
Is it really running hot, or is it a bad thermostat being put on these? The vapor chamber test would be to switch coolers from a known good GPU and slap that sucker on a hot GPU. If that fixes it then it's the vapor chambers for sure!
These are on-die temperature sensors, so there isn't anything to "put on" particularly that would cause the malfunction. Is it possible that the designed-in temperature diodes or sensing logic has some sort of a defect? I don't see how it would pass validation to begin with then, or how there isn't more damage, but i guess it would be silly to dismiss the possibility out of hand either. But seems exceptionally unlikely to me, more likely something with cooler subassembly.
I had issues with high overal temps with hotspot reaching 110C on my rx 6700xt pulse. I even bought morpheus cooler but when I wanted to install it ive realised the standoffs are too long so i packed it back and now the temps are much better with stock cooler. Apparently it was just badly mounted and reseating fixed it. Also ive first screwed cooler to card, gpu core bracket was last to screw in.
TLDR: Outer screws first then springs worked for me
How do you test your card? Horizontal or vertically mounted inside the pc case? Because AMD says if you mount vertically you're fine, but if you mount horizontally as normal in most pc cases then your stock card maybe can overheats. I have no idea why vertical or horizontal matters but that's what AMD says.
I think that might be becuse of the vapor chamber, if there is a kind of over pressure thing and if because that heat transfer with pipes or die starts to goes away this behavior can be happen, and the score issue with ASRock card possibly because of VRAM frequency.
My Sapphire 7900XTX has no issues, temperatures are very similar to the values you got. It's still got the stock thermal paste. I have now applied the auto undervolt setting in the AMD Adrenalin drivers, but it hasn't made a great deal of difference
I just put liquid metal on mine, as if it were a PS5. I took all necessary precautions and now I get 60ºC edge and 70ºC hotspot while playing Horizon Zero Dawn and at maximum settings, of course.
should have tested his card before you changed his paste to verify the issue was still there as he said.
also playing the same game as he showed may have helped reproduce the issue, as it may have
been software related. perhaps some 3rd party app was incorrectly adjust the cards fan speed or disabling
them all together?
I had the same issue with my card. It would reach 110C and I guess it must have went further because it happens two times where my PC completely shutdown to protect the card. I was running the 22.12.1 drivers. Then, they released 22.12.2 and everything seems to be back to normal now with deltas similar to what you have. However, something tells me that probably the thermal paste and pads need to be changed, maybe the pressure if not good enough; comparing my 3dmark results with people who have the same CPU, I found my results to be quite low in the area of 21500-22000 (OC) whereas some people have been getting 27000-29000. If I don't OC the card, the results are even lower, around 19000. Not sure what is going on.
Sorry to hear about your card, looks like the new drivers lowered your gpu clock speed, not good.