Mem Clock Thermal Throttle 3080 FTW (Should I repad?)

Mosie100

Limp Gawd
Joined
Jan 4, 2010
Messages
360
So I've been dabbling and passively mining ETH while I go about my day-to-day operations here on the homefront.
Here are all my settings and results for my 3080 FTW Ultra when I use EVGA Precision :
-150 core, +1000 mem, 68 Power, 70 fan
Average 95 MH/s

Recently my mem temps have been throttling around 110C and then all of the sudden the hashrate drops to 50 MH/s. Really frustrating....
It'll work great for 3-4 days when I'm away for the weekend and then one day it just tanks....
I've thought of everything and tried everything (DDU, MSI Afterburner, reseat GPU to different slot, etc.), EXCEPT a thermal repad.

I've read about the thermal repad and I'm totally spooked about it. I don't want to screw anything up, and have considered just doing an Advanced RMA.
I read this thread here and am about to spend $100 on pads and paste instead of the RMA (thermalright thermal pads and noctua NT-H1 paste).
Does anyone have any thoughts or suggestions on this? I'm really not sure where to go. I'm very comfortable building and taking apart PCs, but have never done a repad and am worried I'm gonna F it up....
Thanks!
 
If you are comfortable enough to remove cooler from gpu, then do it. You just need to get the right thickness pads.
 
If you are comfortable enough to remove cooler from gpu, then do it. You just need to get the right thickness pads.
I feel like such a noob with this. I’m worried I’ll mess the entire thing up
 
I'd say that is possible if you are really inexperienced but unlikely if you are careful and take your time.
I’ve built around 6-7 computers and am extremely comfortable with taking my pc apart and moving things around. Putting the thermal paste on a cpu has never been an issue for me. I just worry about taking apart the gpu for some reason. I’m having a mental block.
Also I’m wondering whether I even need to do it. Or should I just rma it...
 
Best way to overcome a fear is to face it. Take your time. Be deliberate and don’t force or rush anything. It’ll turn out fine. And yes, you should do it.
 
I’ve built around 6-7 computers and am extremely comfortable with taking my pc apart and moving things around. Putting the thermal paste on a cpu has never been an issue for me. I just worry about taking apart the gpu for some reason. I’m having a mental block.
Also I’m wondering whether I even need to do it. Or should I just rma it...
Devil's advocate: RMA the GPU for what exactly? What case will you make to EVGA, when the card runs fine at factory settings and completes benchmarks without error? They won't accept "memory doesn't OC sky-high for mining like some other people that have this card" as RMA reason. 110C mem TJ also isn't RMA-able. If you fib or exaggerate to get an RMA approved ("crashes during benchmarks"), you'd also need to be prepared to possibly get a used/refurb/beater 3080 in exchange as some people have gotten. Maybe the replacement's mem will OC higher, maybe it'll be a lateral move.
 
Last edited:
Devil's advocate: RMA the GPU for what exactly? What case will you make to EVGA, when the card runs fine at factory settings and completes benchmarks without error? They won't accept "memory doesn't OC sky-high for mining like some other people that have this card" as RMA reason. And if you fib or exaggerate to get an RMA approved ("crashes during benchmarks"), you'd also need to be prepared to possibly get a used/refurb/beater 3080 in exchange as some people have gotten. Maybe the replacement's mem will OC higher, maybe it'll be a lateral move.
What I would say: “When I game at regular settings the PC crashes.”
 
What I would say: “When I game at regular settings the PC crashes.”
Worth a shot.

The alternative is buy this to swap out the 2mm VRAM pads only, and call it a day. I've done this mod on a friend's FTW3, you don't need $100 of bullshit detailed in the reddit writeup that you linked- that guy went way overboard for no gain. The pads EVGA uses on the FTW3 are actually decent, and the move 12.8 W/mK pads will be more slight than drastic in TJunc temp reduction.

The mod's pretty easy, like a 5-10min job - taking off the backplate and separating the PCB from the HSF, replacing the VRAM pads, re-pasting the core and putting it back together. I didn't even disconnect any fan connectors. The trickiest part is actually removing the anti-tamper sticker over one of the backplate screws -- it needs to be heated up with a heatgun or hairdryer and removed without deforming it. Edit: I realized that since the Hybrid kit exists for this card, they shouldn't deny an RMA if that sticker's removed.
 
Last edited:
As an Amazon Associate, HardForum may earn from qualifying purchases.
Might be worth a try.

The alternative is buy this to swap out the 2mm VRAM pads only, and call it a day. I've done this mod on this card, you don't need $100 of bullshit detailed in the reddit writeup that you linked- that guy went way overboard for no gain. The pads EVGA uses on the FTW3 are actually decent, and the move 12.8 W/mK pads will be more slight than drastic in TJunc temp reduction. It's pretty easy, the trickiest part is actually removing the anti-tamper sticker over one of the backplate screws -- it needs to be heated up with a heatgun or hairdryer and very carefully separated without deforming it. But taking the card itself apart, replacing VRAM pads, re-pasting core, and putting it back together is fairly easy.
Wow -- thanks for all of that info.
Did that 2mm VRAM pad really resolve all issues with over heating? What was the mem temp difference?

Also -- is it that much of a pain to take off the sticker on the backplate? Was going to get the Noctua NT-H1 paste. Do you recommend that one?
 
As an Amazon Associate, HardForum may earn from qualifying purchases.
Wow -- thanks for all of that info.
Did that 2mm VRAM pad really resolve all issues with over heating? What was the mem temp difference?
In the case I described, there wasn't an overheating issue, as 100C TJunc @ +1000Mhz mem OC was still within spec. The issue was my friend assumed it was TJunc temp alone holding him back from pushing mem past +1000, and I had to break it to him that there's still silicon lottery variance to the GDDR6x that may be contributing. So I did the 2mm pad swap anyway, and TJunc dropped down to 88C, but mem OC still wouldn't push past +1000. That was all he was going to get out of that card. I had a different high-end 3080 that was even worse: TJunc temp 80C-90C, but mem OC wouldn't push past +800 no matter what I tried (crashed) -- just a subpar lottery roll.

And at other end of spectrum, I came across a used, base model Gigabyte 3080 Eagle with a bad reputation, barebones HSF, and worst stock thermal pads I've seen (marshmallow consistency), yet this thing happily ran the mem OC stable at a stellar +1450MHz, even with TJunc at 108C, fans 100%. After a pad swap, TJunc temp dropped to 78C - a 30C reduction - and did it at 75% fanspeed instead of 100%. Bananas. Mem OC became bulletproof at +1500, and it hashes at 102.5MH/s @ 209W. It destroys all other 3080's I've tested including high end Strix, Suprim, FTW3. Silicon lottery.

1617607294763.png
 
Last edited:
Also -- is it that much of a pain to take off the sticker on the backplate? Was going to get the Noctua NT-H1 paste. Do you recommend that one?
It's not that it's hard to take off, it's that it's an anti-tamper sticker that requires extra care to take off and put back without it being apparent. I don't know how militant EVGA is about those stickers at RMA inspection. But like I said, the hybrid kit exists, and would require the sticker being removed for installation, so I assume they wouldn't be dicks about detecting that the backplate had been removed and reassembled.
 
It's not that it's hard to take off, it's that it's an anti-tamper sticker that requires extra care to take off and put back without it being apparent. I don't know how militant EVGA is about those stickers at RMA inspection. But like I said, the hybrid kit exists, and would require the sticker being removed for installation, so I assume they wouldn't be dicks about detecting that the backplate had been removed and reassembled.
Ah, alright well I think I'm going to go through with this. You really think I should just do the VRAM and not the other areas of the GPU?
So basically only in these areas outlined in red I should apply thermal paste and thermal pads? Should I reapply thermal paste on the GPU itself?
vram.jpg
 
the memory is what is overheating, so that is what the priority is. You may also want to put pads between the pcb behind the ram chips and the backplate.
 
the memory is what is overheating, so that is what the priority is. You may also want to put pads between the pcb behind the ram chips and the backplate.
Do you mean here? Sorry, I don't have much experience with the terminology kind of just guessing at this point! lol
here.jpg
 
yeah, do a bit more reading up on the topic before you order pads, specifically search for your model card. But a lot of people are using pads on the backside of 3080s to bridge heat from the back of the ram chips to the backplate as well as putting better pads on the face of the card between memory and heatsink.
 
yeah, do a bit more reading up on the topic before you order pads, specifically search for your model card. But a lot of people are using pads on the backside of 3080s to bridge heat from the back of the ram chips to the backplate as well as putting better pads on the face of the card between memory and heatsink.
Thanks a lot for this help --- I've been reading a ton about it and actually the pads that DPI suggested are the ones I was going to use. Just need to make sure I use the right size, I believe the reddit thread was saying 3mm on the backplate and then 2mm for the ram chips.
 
My suggestion to only bother with the 2mm VRAM pads is based on having actually spent (wasted) the time doing entire re-pads - 1mm, 2mm, 3mm backpads, but then not seeing any gains after 2mm VRAM pads alone. Therefore I won't bother with anything but those in future. But if you want to put the time in and buy multiple quantities of 1mm, 2mm, 3mm, knock yourself out - it won't make things worse if applied correctly.

And yes clean the core with rubbing alcohol and re-paste; it doesn't have to be Noctua paste, but it's standard practice after removing a heatsink from CPU or GPU.
 
Last edited:
My suggestion to only bother with the 2mm VRAM pads is based on having actually spent the time doing entire re-pad jobs - 1mm, 2mm, 3mm backpads, but not seeing any gains versus just doing the VRAM pads. Therefore I won't waste time again on anything but VRAM pads. But by all means if you want to put the time in, knock yourself out, won't make things worse if done right.

And yes re-paste the core because that's a no-brainer, the factory paste is typically subpar no matter which card.
Can't thank you all enough -- going to buy the 2mm pads right now, the Noctua NT-H1 Paste, and some more rubbing alcohol because I ran out :D :D
Thank you I'll post back here with some status updates!
 
This isn't a very [H] answer but you could also just try not hammering your mem for a couple extra MH/s. I get 92 at +500, I personally wouldn't run +1000 just to get 3 more.
 
This isn't a very [H] answer but you could also just try not hammering your mem for a couple extra MH/s. I get 92 at +500, I personally wouldn't run +1000 just to get 3 more.
That's really helpful I'm at +700 now and seems to run steady... What's your core clock at? I'm running -150 and then 68 Power..

Thermal pads come in tomorrow!
 
I run my core at 0 and power at 70.

Common settings I see online indicate I should be able to run lower on both, but lowering either results in a significant hash rate drop for me.
 
I run my core at 0 and power at 70.

Common settings I see online indicate I should be able to run lower on both, but lowering either results in a significant hash rate drop for me.
What's a significant drop for you? I'm getting 95-96 right now but when it all shits the bed it throttles down to like 50 MH/s.

Edit: Just realized you and I have a very similar rig! Sweet.
 
Last edited:
It drops to the low 80s IIRC. Certainly not as bad as the throttling drop you are seeing.

LOL about the similar rigs. I would probably still be running my old 6600k if Microcenter didn't clearance out the 9700ks
 
Wow -- not nearly as bad as mine..

YES - Literally jumped on that 9700k a few months ago for $200 and then $20 off the mobo. A complete steal.
 
I have the same card and my memory temps at least according to precision don't even get over 70c. I run 63% power +900 mem -300 core and i average around 94 mh/s at those settings. I am pretty sure the main part of this cooler shares duty with both the core and the ram so if you are running higher power target and heating things up more its gonna also have an effect on your ram temps due to the core heating up the surrounding block. I would push my memory higher but anything past 900 i tend to get more errors than i would like (more than 1%)
 
I have the same card and my memory temps at least according to precision don't even get over 70c. I run 63% power +900 mem -300 core and i average around 94 mh/s at those settings. I am pretty sure the main part of this cooler shares duty with both the core and the ram so if you are running higher power target and heating things up more its gonna also have an effect on your ram temps due to the core heating up the surrounding block. I would push my memory higher but anything past 900 i tend to get more errors than i would like (more than 1%)
I didn’t know you could even check mem temp in precision. I had to download gpu-z for that. Where do you see the mem temps?
 
I didn’t know you could even check mem temp in precision. I had to download gpu-z for that. Where do you see the mem temps?



When you get into precision you click the "ICX" button . only works on the ICX cards with the extra sensors like the FTW cards

1618153464387.png


1618153609550.png
 
When you get into precision you click the "ICX" button . only works on the ICX cards with the extra sensors like the FTW cards
Ah, I never knew that, thanks! I don't think that's the actual temps though, seems too low... Taking a screenshot now showing miner with EVGA temps and GPU-Z right next to it.
mem temps.png
 
IDK... I would trust EVGA's reading of their own sensors over a third party program that may or may not be reading things correctly. Other vendors don't have this level of sensors
 
I guess the reading in GPUZ is the "junction" temp on the memory and ICX is reading the Die temp or TCASE
 
I guess the reading in GPUZ is the "junction" temp on the memory and ICX is reading the Die temp or TCASE
Yeah I think the junction temp shows when it's throttling -- when it hits like 110 it starts throttling. Absolutely brutal... I'm going to place some thermal pads on the vram in a few days.
 
This isn't a very [H] answer but you could also just try not hammering your mem for a couple extra MH/s. I get 92 at +500, I personally wouldn't run +1000 just to get 3 more.
I run my my 3080 ftw 3 at 60% power -500 core +1500 memory and get 101mh/s I think it is worth doing for 10% more performance. Granted my card is water cooled and memory sits at around 50c per icx readings.
 
I run my my 3080 ftw 3 at 60% power -500 core +1500 memory and get 101mh/s I think it is worth doing for 10% more performance. Granted my card is water cooled and memory sits at around 50c per icx readings.
That is epic! If only I had the water cooling... :)
 
Back
Top