Should I make a thermal maintenance on my GPU?

[mclaren85]

Well I played God of War: Ragnarök for about 30 minutes with ultra settings (DLAA) to measure the temperature of my system. According to latest GPU-Z log data, the temperatures are shown below. Do you think I should change the thermal pads or thermal grease of my 3080? Cause the Hotspots temperature looks dangerously high.

 

[Onasi]

Nah. If you aren’t seeing thermal throttling you are fine and GPU temps themselves are solid. Keep in mind this is Ampere, which was notoriously fairly hot due to the dogshit Samsung node. You aren’t going to get much out of swapping the pads (that’s not helping the hotspot). Swapping the default thermal paste for a PTM pad miiiiiiight improve things a bit, but the difference is unlikely to be massive for the hotspot, that reading is often iffy. Not going to hurt or anything, but disassembling the whole card is always a hassle and WILL require you to change all the pads for ones that are the same thickness, which can be a pain to find out.

 

[Macro Device]

@Onasi, yeah, I'd rather undervolt it.

 

[mclaren85]

Nah. If you aren’t seeing thermal throttling you are fine and GPU temps themselves are solid. Keep in mind this is Ampere, which was notoriously fairly hot due to the dogshit Samsung node. You aren’t going to get much out of swapping the pads (that’s not helping the hotspot). Swapping the default thermal paste for a PTM pad miiiiiiight improve things a bit, but the difference is unlikely to be massive for the hotspot, that reading is often iffy. Not going to hurt or anything, but disassembling the whole card is always a hassle and WILL require you to change all the pads for ones that are the same thickness, which can be a pain to find out.

No never have experienced thermal throttling. Thanks for your help

 

[mclaren85]

This is the furmark test (at 1440p)

After I apply Arctic mx-4 to GPU die, here are the final results:

 

[GerKNG]

you have a suprim x with a 3 digit hotspot?
buy either a PTM from Honeywell or thermal grizzly or some TFX/MX6 (Revision 2 only, the original MX6 is utter trash)

 

[mclaren85]

you have a suprim x with a 3 digit hotspot?
buy either a PTM from Honeywell or thermal grizzly or some TFX/MX6 (Revision 2 only, the original MX6 is utter trash)

Arctic mx4 is fine but only for now. In the long run, it degrades sharply.
I will try to find the original ptm or tfx.. here in my country, thermal pastes and pads are scarcely found

 

[Onasi]

@mclaren85
Heilos PTM pads from Thermalright are also a solid option, if that’s something you can grab more readily available. V2 at least is basically indistinguishable in terms of performance from Honeywell and Grizzly stuff.

 

[eidairaman1]

Your room temperature also dictates how hot something will run, blowing hot air on already hot parts just makes it worse

 

[Lew Zealand]

I would undervolt and repaste.

I did undervolt and repaste. Used 3070 from eBay with dried out paste shot right to 83, 85, then 87C (before I stopped it) with the Hotspot over 105, so smaller deltas than yours. Repaste and run at stock clocks with undervolt plus VRAM OC and it's happy at 66C max Cores and 80C max Hotspot with its so-so Gigabyte Eagle cooler.

 

[eidairaman1]

I would undervolt and repaste.

I did undervolt and repaste. Used 3070 from eBay with dried out paste shot right to 83, 85, then 87C (before I stopped it) with the Hotspot over 105, so smaller deltas than yours. Repaste and run at stock clocks with undervolt plus VRAM OC and it's happy at 66C max Cores and 80C max Hotspot with its so-so Gigabyte Eagle cooler.

Ptm the darn thing

 

[freeagent]

I am using a Heilos pad on my 4070Ti since the summer, so far so good. Right now I have no intention of opening it up, it works well. I might have gone a bit overboard. I just put the whole Intel pad on the heatsink and screwed it together lol. It is for sure making good contact

 

[mclaren85]

@mclaren85
Heilos PTM pads from Thermalright are also a solid option, if that’s something you can grab more readily available. V2 at least is basically indistinguishable in terms of performance from Honeywell and Grizzly stuff.

You mean I should use pad instead of paste?

I am using a Heilos pad on my 4070Ti since the summer, so far so good. Right now I have no intention of opening it up, it works well. I might have gone a bit overboard. I just put the whole Intel pad on the heatsink and screwed it together lol. It is for sure making good contact

Can I use it on CPU as well?

 

[Chrispy_]

Nah. If you aren’t seeing thermal throttling you are fine and GPU temps themselves are solid. Keep in mind this is Ampere, which was notoriously fairly hot due to the dogshit Samsung node. You aren’t going to get much out of swapping the pads (that’s not helping the hotspot). Swapping the default thermal paste for a PTM pad miiiiiiight improve things a bit, but the difference is unlikely to be massive for the hotspot, that reading is often iffy. Not going to hurt or anything, but disassembling the whole card is always a hassle and WILL require you to change all the pads for ones that are the same thickness, which can be a pain to find out.

No never have experienced thermal throttling. Thanks for your help

83c core / 105c hotspot is the default thermal limit, so it *is* throttling. The fact that it's throttling on hotspot rather than core average implies that the paste is absolutely bone-dry and now completely useless.

As long as you're careful when you remove the cooler there should be no need to change the thermal pads on the VRAM - those temps still look fine.

It's sampling every second but you can see after each of the highlighted high temperatures it has a big drop by a whole degree or two followed by a slower climb back up to the 105C throttle limit.

 

[Onasi]

You mean I should use pad instead of paste?

For high wattage and temperature applications like your 3080 PTM is usually seen as better and longer lasting than any paste, yes.

Can I use it on CPU as well?

Yes, they even come in sizes specifically tailored to AMD and Intel IHS dimensions. Not as necessary for CPUs though if you aren’t OCing to pull a lot of power through the chip.

@Chrispy_
Good point, that might indeed be what’s happening. I usually see card hitting the internal BIOS hotspot limit just blackscreen and reboot under heavy load though rather than throttle, so that’s curious.

 

[mclaren85]

83c core / 105c hotspot is the default thermal limit, so it *is* throttling. The fact that it's throttling on hotspot rather than core average implies that the paste is absolutely bone-dry and now completely useless.

As long as you're careful when you remove the cooler there should be no need to change the thermal pads on the VRAM - those temps still look fine.

View attachment 383621

It's sampling every second but you can see after each of the highlighted high temperatures it has a big drop by a whole degree or two followed by a slower climb back up to the 105C throttle limit.

Thanks for the good analysis and the heads up.. The paste was relatively newly applied (mx-4), only 8 months old..
So I repasted yesterday (only I have mx-4) and so far the hot spots temps are 70-75 degree Celsius.

 

[Chrispy_]

Thanks for the good analysis and the heads up.. The paste was relatively newly applied (mx-4), only 8 months old..
So I repasted yesterday (only I have mx-4) and so far the hot spots temps are 70-75 degree Celsius.

Wow, I guess MX-4 is no good for GPUs if it's only 8-month old paste. Did you notice whether it was dried out or completely pumped out?

Either that or there was something wrong with the reassembly when you repasted 8 months ago, and you've solved that issue when you repasted yesterday.

 

[mclaren85]

Wow, I guess MX-4 is no good for GPUs if it's only 8-month old paste. Did you notice whether it was dried out or completely pumped out?

Either that or there was something wrong with the reassembly when you repasted 8 months ago, and you've solved that issue when you repasted yesterday.

I have changed thermal paste of different computers countless of times.. But this time it was so different. The grease was almost completely pumped out. I wish I had taken photo, it was looking so strange and funny; I would say %90 of the grease gathered outside of the chip. Yet it felt it was not dried out.

I will continue to monitor the temp though. Not sure if it will work like this for long.

 

[kiddagoat]

This is just my opinion, I have experienced similar issues in the past on GPUs.

I would use a thicker paste such as TF9/TFX or one of the newer pads available such at PTM or Kryosheet.

I was using MX-2 and recently some MX-6 on the bare dies and it just seems to not last as long as it used to.

So I still use the MX-6 IHS and TFX on bare die. YMMV though.

 

[mclaren85]

This is just my opinion, I have experienced similar issues in the past on GPUs.

I would use a thicker paste such as TF9/TFX or one of the newer pads available such at PTM or Kryosheet.

I was using MX-2 and recently some MX-6 on the bare dies and it just seems to not last as long as it used to.

So I still use the MX-6 IHS and TFX on bare die. YMMV though.

You are definitely correct according to other sources and user experiences. I've learned it in a hard way:/
ps. kryosheet is a good option but electrically conductive..

 

[eidairaman1]

This is just my opinion, I have experienced similar issues in the past on GPUs.

I would use a thicker paste such as TF9/TFX or one of the newer pads available such at PTM or Kryosheet.

I was using MX-2 and recently some MX-6 on the bare dies and it just seems to not last as long as it used to.

So I still use the MX-6 IHS and TFX on bare die. YMMV though.

It appears the in thing is to use thermalpads whether ptm like or carbon like.

 

[freeagent]

Can I use it on CPU as well?

Yessir.

The grease was almost completely pumped out.

Not at all surprised really. I don't use that stuff.

 

[mclaren85]

Look what happened after the paste:

I hope the previous temp values didn't harm my gpu, isnt it?

 

[Space Lynx]

Look what happened after the paste:

View attachment 383911

I hope the previous temp values didn't harm my gpu, isnt it?

should not have harmed anything no, but its good you got such better results with the repaste. at least you got it sorted now. probably won't need to worry about it for 3 years or so, is my guess.

 

[Lew Zealand]

Look what happened after the paste:

View attachment 383911

I hope the previous temp values didn't harm my gpu, isnt it?

You're probably fine. My used 3070 was probably running over 90C for ??? time with the previous owner before they sold it and works like champ for me after a repaste. Was there some silicon degradation in there? Maybe, but for $220 it's great and I don't run it at max except in benchmarks. The rest of the time it's chillin' at ~170W and 60C max.

Your GPU clock is super low there, looks like you may have gone overkill with the undervolt/underclock. For reference, my daily driver MHz on the 3070 is 1875MHz at 0.893V. Even with a conservative clock and UV like I like to run, you should be in that general range depending on silicon lottery. I only see you hangin' at a low ~1000 MHz in that chart.