ASUS G531GW: noticeably higher GPU temperatures after repasting

[R0mst3r]

Hi everyone!

So I decided to change thermal paste in my ASUS ROG Strix SCAR III (model № G531GW, i7 9750H + RTX 2070, 16 GB RAM), as the last repasting was done by an Authorized Service ~14 months ago during the faulty cooling fan replacement.
Did some research by reading forums, watching some YT videos on the topic, and decided to go with Arctic MX4 as it was universally loved and recommended by a lot of people who own this exact laptop model and other fairly hot ASUS ROG laptops.

I had some doubts about what to do with the thermal goo put by ASUS on video memory chips and VRM elements, as you obviously can't buy that exact stuff ASUS used during the manufacturing proccess. Some people advised to clean the goo off and use the same thermal paste as on CPU and GPU crystals. Fewer people were in favor of replacing the goo with thermal pads, but the recommended thickness of the pads was really controversial topic.

When I disassembled my laptop, I saw that the thermal goo was pretty dry and didn't really cover the surface of video memory chips and VRM elements evenly, there were uneven clumps of it on some chips, some of them were completely clean, and the overall 'plasticity' of the goo didn't let me to reapply it evenly (I honestly tried). Then I decided to clean it off and to use thermal paste instead as was recommended by some folks on the web. I didn't know how thick the layer of paste should be where the goo were previously, so I went with the same fairly thin layer as on GPU and CPU crystals.

After assembling my laptop and switching it on I noticed an improvement in CPU temperatures right away. Before the repasting it was 55-60° С in idle in Silent mode (the one in Armoury Crate), after the repasting it was 48-52° in the same conditions.
But as soon as I tested it in games, I noticed a problem with GPU temperatures. Before the repasting, GPU was never hotter than 76-77° C under the max load (e.g. Cyberpunk 2077 with RTX lightning 'on' utilizes 97-99% of GPU resources according to GeForce experience overlay, and I could play it for unlimited amount of time with GPU temps staying at 76-77° C). After the repasting, GPU now gets so hot that it easily reaches 86° C in a couple of minutes in game (Cyberpunk 2077, Battlefield 3) and that is the moment it starts to throttle, reducing the FPS for a couple of seconds to drop the temps of GPU to 80° C, and then it goes up to 86° C and throttles again, the cycle never ends as long as any 3D game is running.
For 2.5 years of owning this laptop I never encountered GPU thermal throttling, so it definitely is a consequence of the repasting.

Now, the question is what exactly did I do wrong and how can I fix it? I definitely tighten all the screws of the heatsink evenly and firmly (but not too tightly), I cleaned all the dust and lint from the cooling system, the cooling fans are working as well as they did before the repasting.

My guess is that it's thermal paste applied instead of thermal goo on the video memory chips causes this problem (there is no separate sensor for video memory in this laptop AFAIK, so the GPU temps sensor probably shows avg of GPU and video memory temperatures, but correct me if I'm wrong about it).
Maybe I should have had a thicker layer of paste where the goo was? But how thick should it be? Maybe I should try some other paste? Or maybe I should buy thermal pads (again, how thick should they be?) and use them instead? Or maybe MX4 is just not that effective for the RTX 2070?

I'm not sure about anything anymore, so maybe someone had the same experience and knows how to solve this issue?

Thanks in advance.

[Murph_9000]

The thermal paste used for CPU cooling is designed for where there's already hard direct contact between the cooler and heat spreader. Its purpose is to fill the imperfections between the two surfaces and create a thermal bond that's essentially 100% of the surface area of the CPU heat spreader, but as an extremely thin layer.

It's not designed for cases where there's an actual gap that needs to be filled. There are a lot of different pastes on the market, so it's possible that some CPU paste may be thick enough and have the correct thermal properties to be used in both scenarios (hard direct contact, and bridging a gap between component and cooler). In particular, some high performance CPU thermal paste is designed to have a fairly low viscosity to enable it to form the thinnest possible layer between CPU and cooler; that type of paste is likely to be particularly unsuitable to fill a gap. CPU paste that's electrically non-conductive can typically be used on the main GPU die, where there's also hard direct contact.

I can't really suggest what the right stuff is for laptop GPU memory chips, if there's a gap to fill, as it's just not an area of expertise for me. The above is just general observations about what CPU paste is designed for. I thought GPU memory was typically thermal pads, but laptop cooling is a special case and can very much be down to exactly how the cooling solution is designed for the specific model.

You could maybe do a bit of trial and error with thermal pad thickness; buy several different thicknesses of pad and see which performs best. You probably don't want something that's far too thick, but the typical thermal pads seem to compress and deform quite well, so slightly thick is probably the way to go to (better than slightly too thin).

[wyvernone]

I have a similar experience. I bought a second hand 2 year old unit and on the first test the CPU was 92+ deg C while GPU was around 73 deg C. Blimey the CPU was running way too hot I thought.
But I should have read this laptop model's reviews first so that I would have known that this is to be expected for CPU to run this hot, and not to bother repasting, even for brand new units sent to reviews at famous laptop reviewers.

Anyway, I forged ahead with the repaste and on the first attempt, I used 1.5 mm (thickness as recommended in a yt video) thermal pads on the VRAMs and power stuff and used Noctua NT-H1 thermal paste (previously I used "diamond" thermal paste but I thought I'd try a different one this time) on the GPU and CPU. After this, the GPU was way too hot it was throttling itself.
On the second try I though maybe I used to much paste on the GPU/CPU and thinned them up but no improvements.
I also identified that when I removed the heatsink assembly I used a little bit too much prying from the fan sides instead of from the centre and this may have put a bit of a curve on the heat pipes - I straightened them out as best as I could but who knows it may have pushed the copper heatsink pad beyond the square/level face-to-face alignment with the GPU die.
On the third attempt I inspected and found the thermal pad is way too thick it prevents the gpu copper heatsink from making contact with the GPU die. I removed all thermal pads on the VRAMs and used thermal paste on them just like the GPU/CPU.

After this GPU temp is about +3 deg C above "before" state running Superposition 1080p extreme benchmark, while CPU seems to be also warmer, such that on Prime95 stress test (without AVX) it hits 98 deg C.

Hmm... I thought this could be ambient temp difference (I didn't record the ambient temp for before /after) and having read the reviews now that say CPU hitting 95+ is normal for this laptop model I am not sure if it can be reduced further although one user has tried liquid metal on his laptop and posted it at

G531GW - Asus Scar 3 / Asus Scar III Liquid Metal repaste - a noob attempts.

If your heatsink assembly is bent a replacement can be purchased from https://www.asus-accessories.com/laptop/cooling-system/heatsink/serie/g531 or from aliexpress.

I ran the RTX demos and found the temps to be decent:
reflections rtx demo, 72 deg cpu, 94 deg cpu, cpu power 79w
justice rtx demo 74 deg gpu, 96 deg cpu
atomic hearts demo 73 deg gpu, 98 deg cpu, cpu power 79w

The CPU drawing 79W is a on a high side and can be resolved with intel XTU by under-TDP the CPU so I am happy with it.

I would recommend @R0mst3r : try a different thermal paste, use the guide Best Thermal Paste for CPUs 2022: 90 Pastes Tested and Ranked to help.
Before putting on the paste and pads, put the heatsink assembly over the cpu/gpu and make sure the heatsink is flushed and level with the cpu/gpu .
Use thermal paste on the vrams, thermal pads 1.5mm on the power components, and you know the rest. Good luck.

[R0mst3r]

wyvernone wrote:
I would recommend @R0mst3r : try a different thermal paste, use the guide Best Thermal Paste for CPUs 2022: 90 Pastes Tested and Ranked to help.
Before putting on the paste and pads, put the heatsink assembly over the cpu/gpu and make sure the heatsink is flushed and level with the cpu/gpu .
Use thermal paste on the vrams, thermal pads 1.5mm on the power components, and you know the rest. Good luck.

Thanks for the reply.
Since my previous post, I already tried applying thermal pads both on video memory chips and VRM elements, went with 0.5 mm Arctic pads (not the best choice, I know).
Why 0.5 mm? Because I found some info in the comments section under one YT video, where some guy said he contacted ASUS about buying the thermal goo they use in these laptops and ASUS said you can't buy the goo, but recommended using 0.2 mm thermal pads instead.
Couldn't find 0.2 mm pads in my country, so I went with 0.5 mm ones.

And after applying the pads I saw some improvements in terms of temperature. GPU doesn't trottle anymore (no FPS dips after reaching 86° C), but the temperatures are still higher than they were before the whole repasting thing. I got stable 85-86° C in Battlefield V, 82-83° C in Battlefield 3 (11 y.o. game), same 85-86° C in Cyberpunk 2077. And there wasn't any thermal throttling despite reaching the 86° C point, which was the throttling point before I applied the pads.

Curious thing to mention: the GPU clocks as high as 1950 MHz according to GeForce Experience monitoring and GPU-Z logs. I honestly don't remember seeing those clocks before the repasting.
So I tried to lock GPU clock in base values using MSI Afterburner, and even with base clock the GPU temperature was still as high as with 1950 MHz clock.
I don't know if it proves anything, but my guess is that this heat originates from video memory chips.

After that, I disassembled my laptop again to check the contact between the pads and memory chips. The contact is very good, I could even see the text engraved on the surface of the memory chips imprinted in the thermal pads.
Also, I was well aware that you can easily bend the heatsink if it's pulled out incorrectly, so I was careful every time. Judging by the looks of it, it isn't bent, so that theory can easily be withdrawn.

The only thing I could try now is to remove thermal pads and apply thermal paste again, but this time just use more of it.
Because last time I went with thermal paste, not a single video memory chip and not a single VRM element contacted the heatsink well-enough to leave some of the paste on the heatsink (it was squeaky clean in those places when I removed it to apply the thermal pads).
I have hopes this will work as I stumbled upon a forum thread (not on ASUS forums) where the guy got rid of thermal goo and had the same problem with overheating, tried thermal pads and K5Pro (looks like an analogue of ASUS'es thermal goo) with little to no success, but after that some guy in that thread recommended to use a little more thermal paste on video memory chips and VRM, and the OP came back and said it was a success, the temperatures went back to the pre-repasting values.

Will be back with an update when (or if) I do that.

[wyvernone]

R0mst3r wrote:


Curious thing to mention: the GPU clocks as high as 1950 MHz according to GeForce Experience monitoring and GPU-Z logs. I honestly don't remember seeing those clocks before the repasting.
So I tried to lock GPU clock in base values using MSI Afterburner, and even with base clock the GPU temperature was still as high as with 1950 MHz clock.
I don't know if it proves anything, but my guess is that this heat originates from video memory chips.

The GPU applies boost if sees any safe headroom, until one of the limits are reached : temp, power, or volt, and then it will downclock itself to normal rates and then go again if the GPU has work in the queues and there is room to boost.

Also remember the heat sink assembly has a shared/common heat pipe between the GPU and CPU so if the CPU runs hot, the heat will automatically seek cooler places and will hit the GPU.

Strange you can't lock in the GPU with MSI Afterburner. Have you tried Ctrl-F to adjust the voltage-frequency curve? That should lock it in.

[R0mst3r]

So, a year has passed since I started this topic, and I have some updates (maybe someone who struggles with a similar problem will find it useful): 

The trick with 'solid' (aka normal, aka 'real') thermal pads on the VRAM and VRM kinda worked. Since I applied them almost a year ago I've been using my laptop quite extensively (~1100-1200 hours of gaming) and while the GPU temperatures were disturbingly high, they never exceeded 86° C. By the way, in my previous post I said that this GPU throttles at 86° C, while in reality the temperature limit is 87° C (according to GPU-Z), that explains why it could run at 86° C for hours and hours. However, that is not the point.

All this time I've been trying to research the topic of 'liquid thermal pads' — that is a proper name for the substance I called 'thermal goo' many many times in this topic. And as a result I found out that there are a couple of options that can be purchased by a casual individual such as myself and applied fairly easily (not really). These options were Laird tPutty 607 and Honeywell HT10000. In reality, there are a lot more options of liquid thermal pads out there, but in my war-torn country (i'm from Ukraine) I managed to find only these two.

Long story short, I bought some Laird 607 and decided to make things right this time. The application process of Laird 607 is tricky as it refuses to adhere to surfaces of chips so one should be very careful while putting it so it doesn't fly all over the motherboard. After holding my breath repeatedly for a good 10 minutes I managed to apply it fairly decently on all 8 VRAM chips and billion VRM elements that should have liquid thermal pads on top of them. For a thermal paste I decided to go with Noctua NT-H2 this time as I've heard a lot of good things about this thermal paste from gaming laptops owners. 

So, the result of all these manipulations was underwhelming to say the least. Don't get me wrong, there are some noticeable improvements, but nowhere near the 'new straight from the factory laptop' temperatures.
After applying Laird 607 and Noctua NT-H2, I ran the Unigine Heaven 4.0 benchmark for almost an hour, and the average GPU temperature was 83.93° C (lets call it 84° C). The max temperature was 85.3, but it stayed that high for maybe 1-2 min during and hour-long session. 
I also tested the temperatures in Battlefield V, and during one TDM round temperature slowly risen from 78° C to 83° C. I'm yet to test the temperatures in Conquest (larger maps, probably higher GPU load), but it's already an improvement over a 'solid thermal pads'-scheme I've been using for the last year, because with solid thermal pads on VRAM and VRM it used to get as hot as 85-86° C instantly upon starting the match, while now it slowly rises from 78 to 83 during a 15-20 min game session. UPD: Played one round of Conquest on Pacific Storm map, this time around the GPU almost instantly heated up to 83° C at the start of the round and after a 5-10 min it was already a stable 86° C. There was no GPU throttling, so I guess it stayed at 86° C and didn't go any higher.Still, this result is the same as it was with solid thermal pads on VRAM and VRM, the only difference is with liquid thermal pad it took some time to heat up to those temperatures, while with solid ones it was at max temps almost instantly.
Cyberpunk 2077 (medium-high video settings, RTX on but only Lightning and only on Medium) during a 13 min session of driving through the Night City heated the GPU up to 85° C, this time the GPU reached the temperature of 83° C almost instantly after loading into the game. I think if I would continue to drive for 5-10 minutes more, I could easily reach 86° C (and 87° C is temperature limit=throttling). My guess is RTX-on setting fires up those rtx-cores inside GPU which Unigine Heaven 4.0 can not utilize thus warming it even more then in non-RTX conditions. I'm yet to test some other games I play frequently, but I doubt it would be any different from the results mentioned above.

In conclusion, I expected more from Laird 607 as it is universally praised on the Internet as one of the most effective liquid thermal pad (better then K5 Pro for sure). To be precise, I expected GPU temperatures to not exceed 80° C during the highest loads with RTX and all other fancy things my RTX 2070 supports. And 76-78° C MAX would be an ideal result for me, but alas — not even close. Although, it's noticeably better compared to applying solid thermal pads and, god forbid, thermal paste on VRAM and VRM. 
Granted, there's a chance I could do something wrong, apply to much or too little of Laird 607, even though I strongly doubt that I did something wrong. We'll see, maybe next time I will go with Honeywell HT10000 instead of Laird 607 (it's 10 W/mK compared to Laird 607's 6.4 W/mK) as well as Honeywell PTM7950 thermal pads with phase-change properties instead of liquid thermal paste on GPU and CPU chip crystals. 
If you have any questions regarding my 'journey' feel free to ask them.

[izy]

Try upsiren ux pro instead of laird it is an extremely well performing thermal putty much higher than the ones you mentioned.

There are also different types of liquid metal like conductonaut extreme.

Liquid metal performs worse if you can't get a solid contact with the CPU which is why putty helps because it smushes down.

Also if you applied LM to a CPU with a copper heatsink it will need reapplied. The copper is porous and will have absorbed some of the lm to fill it's own micro gaps. You will notice it is permanently stained silver now this is fine.

If reapplying doesn't improve at all it will definitely be a contact issue.

Other than that replacing it with Honeywell ptm 7950 instead of lm could give results since it is thick for contact .

[R0mst3r]

---

@izy wrote:

Try upsiren ux pro instead of laird it is an extremely well performing thermal putty much higher than the ones you mentioned.

There are also different types of liquid metal like conductonaut extreme.

Liquid metal performs worse if you can't get a solid contact with the CPU which is why putty helps because it smushes down.

Also if you applied LM to a CPU with a copper heatsink it will need reapplied. The copper is porous and will have absorbed some of the lm to fill it's own micro gaps. You will notice it is permanently stained silver now this is fine.

If reapplying doesn't improve at all it will definitely be a contact issue.

Other than that replacing it with Honeywell ptm 7950 instead of lm could give results since it is thick for contact .

---

Thanks for the advice. I'll make sure to consider upsiren ux pro the next time, 16.8 W/mK sounds huuuuge (if true).

Not sure if I'm ready to deal with LM, from what I've seen the process isn't that easy + there's a huge risk of electrical damage if done improperly. And I don't think I know repair shops which I trust enough to do it for me 'professionally'. 

Also, the more I think about it, a contact issue could really be the case. This time around when I lifted the heatsink I noticed that thermal paste hasn't spread that well on the GPU crystal and the corresponding heatsink area, like it has on CPU. Here's the photo (don't mind the solid thermal pads, I got rid of them): 

I don't really know if that's normal paste spreading or there's something wrong with the contact on GPU side.

[izy]

Yeah looks like uneven pressure/spreading on GPU since paste has shifted left

[R0mst3r]

---

@izy wrote:

Yeah looks like uneven pressure/spreading on GPU since paste has shifted left

---

Is there any 'easy' way to fix it without buying a new heatsink?