Question Zotac RTX 3090 Trinity OC - - - instant 110c memory temp ?

[kapul4]

Hello, I bought 2nd hand rtx 3090. It has been repadded (idk which type of pads, I see they are blue). 800mv undervolt/1750mhz makes it stable, with corr temps while benchmarking or using stable diffusion never going above 65-70c at 2000rpm fan speed, but after like 20 seconds of intense load the memory temp goes to above 100c and slowly creeps up to 110c, starting to throttle.

Reducing memory clock to maximum of -500 in afterburner makes no difference, as does reducing power limit to 80%. Mem temp will slowly reach 110c. While idle, it stays the same temperature as the core. Increasing the fan speed to 100% does very little to the mem temp so I think guy used the wrong thermal pads. Is there a way to just remove the backplate on the zotac card without removing the whole cooler(would have to reaply thermal paste) in the process to change the thermal pads. Also is there a way to undervolt the memory?

 

[thestryker]

There's basically one of two likely things at work: you got a card that was used for mining and the VRAM was damaged, or all of the thermal pads need to be replaced (meaning VRAM on front and back as well as VRMs). Just make sure you get pads with the correct thickness.

 

[kapul4]

There's basically one of two likely things at work: you got a card that was used for mining and the VRAM was damaged, or all of the thermal pads need to be replaced (meaning VRAM on front and back as well as VRMs). Just make sure you get pads with the correct thickness.

Will thermal grizzly chill pad 8 work with 2mm thickness? The internet is biased on the thickness. I dont see how memory damage can result in increased temperature tho.

 

[thestryker]

I'm not sure whether it will or won't as I've never used that type of pad. All I saw on a quick search was that some people were able to use 2mm on both sides just fine and others used 3mm on the back. 2mm should be fine as I'm assuming those who used 3mm got very soft pads and I've never used those TG ones so don't know how soft they are or aren't. I saw nothing regarding VRM pads so they may be the same thickness.

Damaged VRAM can have the voltage regulation off or just be getting more power pushed to them to operate.

 

[kapul4]

I'm not sure whether it will or won't as I've never used that type of pad. All I saw on a quick search was that some people were able to use 2mm on both sides just fine and others used 3mm on the back. 2mm should be fine as I'm assuming those who used 3mm got very soft pads and I've never used those TG ones so don't know how soft they are or aren't. I saw nothing regarding VRM pads so they may be the same thickness.

Damaged VRAM can have the voltage regulation off or just be getting more power pushed to them to operate

I took the cooler apart today, if there is not much imprint on the thermal pad, it means it's not making good contact? Mem power peaks at about 140w

 

[kapul4]

Off topic, is there an aio for zotac trinity? I couldnt find specific info on google

 

[Phaaze88]

The internet is biased on the thickness.

Because pad hardness(for example, soft/medium/firm) is a variable among them that gets neglected(I don't know why) and all the different manufacturers don't have a universal standard for testing their pads. It's often, 'thickness this, thickness that', and when it doesn't work out, the user is stuck between a rock and a hard place, as they don't know where to go from there.
Another problem is that the hardness isn't directly stated to shoppers.
Best to leave pads alone, unless one absolutely has to, but that doesn't seem to be the case for you.

is there an aio for zotac trinity?

Alphacool Eiswolf 2.

KOI Alphacool Eiswolf 2 AIO - 360mm RTX 3090/3080 with Backplate (Reference) PHT EOL

Fullcover all-in-one water cooler for the Geforce RTX 3080/3090 - 360 mm copper radiator - Expandable - ARGB

shop.alphacool.com

 

[thestryker]

I took the cooler apart today, if there is not much imprint on the thermal pad, it means it's not making good contact? Mem power peaks at about 140w

Generally speaking yes that does mean it's not making good contact, and for sure is the case if the pad is soft.

 

[kapul4]

Because pad hardness(for example, soft/medium/firm) is a variable among them that gets neglected(I don't know why) and all the different manufacturers don't have a universal standard for testing their pads. It's often, 'thickness this, thickness that', and when it doesn't work out, the user is stuck between a rock and a hard place, as they don't know where to go from there.
Another problem is that the hardness isn't directly stated to shoppers.
Best to leave pads alone, unless one absolutely has to, but that doesn't seem to be the case for you.


Alphacool Eiswolf 2.

KOI Alphacool Eiswolf 2 AIO - 360mm RTX 3090/3080 with Backplate (Reference) PHT EOL

Fullcover all-in-one water cooler for the Geforce RTX 3080/3090 - 360 mm copper radiator - Expandable - ARGB

shop.alphacool.com

I have seen that alphacool aio before, but wasn't sure does it fit the zotac since it says its for reference gpu. not sure does the zotac have the same board layout. the pads are rather firm, there is an outline of the memory chips on the pad but not how i remember some other cards having them inprinted into the pad. also it seems they are also 2mm in thickness

Edit: mem temps dropped to 96-100c doing stable diffusion with a passive backplate with fins i had left over from an old arctic accelero twin turbo cooler from gtx 580 days. "it may come in handy" hits different now

 

[Phaaze88]

I have seen that alphacool aio before, but wasn't sure does it fit the zotac since it says its for reference gpu.

I entered your card in Alphacool's Gpu Configurator, and if you scroll down to 'Compatibility' in the link, you'll see your model listed there.

not sure does the zotac have the same board layout.

http://gpu.watercool.de/WATERCOOL_HEATKILLER_GPU_Compatibility.pdf

The Trinity uses the reference PCB, but it's an elongated form.

 

[kapul4]

I entered your card in Alphacool's Gpu Configurator, and if you scroll down to 'Compatibility' in the link, you'll see your model listed there.

http://gpu.watercool.de/WATERCOOL_HEATKILLER_GPU_Compatibility.pdf

The Trinity uses the reference PCB, but it's an elongated form.

So it fits, awesome. Tnx

 

[kapul4]

Generally speaking yes that does mean it's not making good contact, and for sure is the case if the pad is soft.

Here is the pic of the pads

 

[kapul4]

I entered your card in Alphacool's Gpu Configurator, and if you scroll down to 'Compatibility' in the link, you'll see your model listed there.

http://gpu.watercool.de/WATERCOOL_HEATKILLER_GPU_Compatibility.pdf

The Trinity uses the reference PCB, but it's an elongated form.

Sadly it wont be in stock for at least a month as they say

 

[kapul4]

UPDATE: Yep, it's ebay returning time. Just 3 days after receiving it... Well better than 30+ days. Sadly the GPU is crashing during Stable Diffusion workloads now, let alone Furmark which fails in like 10 seconds if the GPU is at stock clocks with no undervolt. Screen simply goes blank, with event viewer reporting no comunication from the GPU with the driver and the pc restarts. Is this the cause of defective memory, the core, or voltage regulators? I am just curious. Can that be a pinpoint was the card used for mining or not?

 

[thestryker]

Mining only tends to be hard on the memory itself as the GPU is typically set for optimal operation efficiency wise. So any power systems etc that feed memory are also going to see more wear. I think it's a safe bet that if you got a card that had new thermal pads which aren't the correct ones it probably was not well taken care of and most certainly wasn't under normal usage. I replaced the pads on my video card with thermal putty, but if I were to sell it I'd have left the putty there.

Buying anything 3080+ used is unfortunately a crap shoot with regards to whether or not it was used for mining.

 

[kapul4]

Mining only tends to be hard on the memory itself as the GPU is typically set for optimal operation efficiency wise. So any power systems etc that feed memory are also going to see more wear. I think it's a safe bet that if you got a card that had new thermal pads which aren't the correct ones it probably was not well taken care of and most certainly wasn't under normal usage. I replaced the pads on my video card with thermal putty, but if I were to sell it I'd have left the putty there.

Buying anything 3080+ used is unfortunately a crap shoot with regards to whether or not it was used for mining.

Mining only tends to be hard on the memory itself as the GPU is typically set for optimal operation efficiency wise. So any power systems etc that feed memory are also going to see more wear. I think it's a safe bet that if you got a card that had new thermal pads which aren't the correct ones it probably was not well taken care of and most certainly wasn't under normal usage. I replaced the pads on my video card with thermal putty, but if I were to sell it I'd have left the putty there.

Buying anything 3080+ used is unfortunately a crap shoot with regards to whether or not it was used for mining.

My friend had 3070 and after a week of buying the card it was reporting 0mb gpu memory, next time the pc wouldnt post. Whats the usual thing that fails on rtx 3080+ cards? Memory? Vrm? Do high tier cards like strix or aorus extreme last longer?

 

[Phaaze88]

My friend had 3070 and after a week of buying the card it was reporting 0mb gpu memory, next time the pc wouldnt post. Whats the usual thing that fails on rtx 3080+ cards? Memory? Vrm?

Stryker is partially right - mining activities beat the crap out of the memory controller in the gpu die. It's not like gpus that were only used for gaming are out of the woods though. Those beat up the VRMs. So, either way...
Buying gpus 2nd hand is a wild west.

Do high tier cards like strix or aorus extreme last longer?

Negative.

 

[kapul4]

UPDATE:
Ok, this is simply OUTRAGEOUS and must be a prank from someone. So, I took the card apart since I had nothing to lose, the new thermal pads were on its way already anyway, partial refund(80% of the price) I agreed to has been issued on ebay and the new RTX 3090 aorus extreme with very good price and 1 year warranty from a PC shop sounds pretty good.

Took the card apart, no damage on the PCB, no bulging caps, all the memories have the same date code and so on. Good amount of thermal paste, all the thermal pads were nicely cut and nothing out of the ordinary, tho I noticed that the MOSFET driver(small 4x4 chip) is missing it's thermal pad, comparing my thermal pad layout to the factory one, it wasn't supposed to be there to begin with (tho I saw people mentioning on the forums to other ppl repading they should put the pad on that small chip) so no wonder the previous guy didn't put it on also. What else was strange to me was, why is there a QR code sticker on the heatsink area that is touching the GPU and the memories? Upon closer look, I noticed that the QR code sticker(which I couldn't read with the phone) is covering half of the solo memory module which is at the bottom of the card! Searching for pictures on the net, it seems that the exact same sticker is there on every Trinity cooler. Could that be causing high memory temps for me and other people? No idea. I peeled it off, cut off the new thermal pads (Thermal Grizzly Minus 8 2mm) with scissors in a much uglier way than the pads already on the card and applied them to the memory and that driver chip, applied new MX-4 thermal paste and put it all back together one last time before selling the card for parts or sending it in for correct diagnosis if possible.

Fired up all the usual programs where the card would skyrocket the mem temps and later on crash with "Event ID 0 from source nvlddmkm" ( remember, it would crash no matter the memory or temperature, it wasn't temperature dependent in short). Guess what?
45 minutes in, mem temps are at 90 to max 104c, gpu at max 70 and hotspot at 80, with both core and mem controller load at 100% but most importantly- Still. Did. Not. Crash.
Could it be that the MOSFET driver was overheating or that the sticker was causing the memory to run at 110c? I doubt a pad change to a mediocre Minus 8 with the same 2mm pad thickness(tho, this ones being softer) as the previous blue pads the guy before me put resulted in no crashing and 10-15c lower memory temps.