I'm expecting mid 80's instead of mid 90's. Think I'd have to do both sides of the card to see the 15 to 18 degree drop some are seeing. 

We shall see. Pad thickness is vital as is pad squishyness on the Shaw Scale. Seems there's more to pads than you might think. Thermal conductivity, hardness and then there's thermal putty as well as pads on the fan side of the board. Annoyingly all third party board companies use diferant thickness pads and putty too. But 2.5mm seems about right for the FTW3 backplate.

Internal case temp and air flow has a big factor in your hardware GPU and CPU temps, I never use the fans supplied with the case but upgrade them to Noctua High performance 140mm this adds to the cost of the build but my GPU fans never spin up unless I get into very heavy scenery load and my CPU stays below 68deg.

54 minutes ago, G-RFRY said:

Internal case temp and air flow has a big factor in your hardware GPU and CPU temps, I never use the fans supplied with the case but upgrade them to Noctua High performance 140mm this adds to the cost of the build but my GPU fans never spin up unless I get into very heavy scenery load and my CPU stays below 68deg.

 

Yep, most definitely.

Case airflow isn't the issue here though, although it can impact DDR6X VRAM temp to a degree. My case airflow is optimal. I've even directed a case fan at high RPM directly at the card, barely any change in VRAM temp. This VRAM Junction Temp issue is well known. You will find all manner of tech YouTubers and owners of the 3080/3090 cards talking about it. And plenty of pad replacement activity going on. Many attach extra large heat sinks to the back plates and attach coolers, makes a slight difference but not much.

As for replacing case fans, that's the first thing I do, and like you I opt for Noctua. Chromax black NF-A14 to be precise.

 

Not just founders edition as per Jayz video below. The vast majority of 3080 and 3090 cards are affected. 

 

 

 

Edited February 17, 20224 yr by martin-w

IN this bench the 3080 runs hotter.

https://www.youtube.com/watch?v=7dofNU0r9D8

 

@martin-w How soon until you are doing the pad replacement?

1 hour ago, G-RFRY said:

IN this bench the 3080 runs hotter.

https://www.youtube.com/watch?v=7dofNU0r9D8

 

 

You don't understand the issue. Its nothing to do with GPU core temperature. My GPU core temp is in the 60's under load.

The issue is VRAM Junction Temp. Its only since the newer versions of Hardware Info and GPUz appeared that monitor junction temp that it became apparent.

Most 3080 and 3090 owners are seeing perfectly normal GPU core temp but VRAM temp that can be between high 80's and 100's. Crypto Miners are seeing throttling. 

EVGA told me they arent bothered with VRAM in the 90's, as long as it doesn't reach 110 and throttle. Which is fine in terms of lasting the warranty period, but at this stage we have no idea regarding VRAM longevity beyond that. 

You would need to read the rest of this thread.

Edited February 17, 20224 yr by martin-w

18 minutes ago, micstatic said:

@martin-w How soon until you are doing the pad replacement?

 

Tuesday probably. Pads come on Monday. I'm only doing the backplate pads as it seems that's the main issue from what I've researched. All that cools them are less than efficient stock pads and the thin aluminium backplate.

 

 

40 series rumored to release in second half of this year, with the 4090 having 30gb of DDR6X.

Well it seems to me that they will need active backplate cooling. Either that or DDR6X that generates less heat.

On 2/17/2022 at 1:31 PM, micstatic said:

@martin-w How soon until you are doing the pad replacement?

 

Just done it. Just the pads on the backplate. With Gelid Ultimate.

Result... ZERO reduction in temperature. 🤣

Not sure if it would make any difference doing the pads on the GPU side of the card, some say yes, some say not much difference.

No idea why Mr Turboken has such low junction temps. After scouring the internet I've come across nobody with such low junction temp unless they have a custom loop with active backplate cooling, or have replaced the pads, or have done the "copper shim" mod. 

Anyway, Micron have updated their website with latest info on DDR6X. They say "operating temp" is 0C to +95C / +105C

I see 94 under load in the sim. 90 in games. Which is what most people are seeing. 

 

 

Edited March 1, 20224 yr by martin-w

@martin-w Thanks for the update.  oh well!

When Micron say 95/105 I'd like to know what they mean. Which is the max 95 or 105. 

Anyway, I've sent them an email, will be interesting to see what they say. Or if they bother to reply at all.

They used to state 95, for some reason they updated it and added 95/105. Not really clear what they mean. And I suspect they are referring to case temp not junction temp. 

Its not a big deal, I suspect that the cards will chug along at 95 depress for an acceptable duration without failure but they really should clarify the situation. 

It would be nice to see if anyone else could test this on there 3090 for comparison. I guess its safe to say I have a well assembled card. Which is good, because at its cost it will likely be in my machine for several more years. Sorry you didn't get the improvement we were all hoping to see Martin. Maybe you should attempt to remove the other side and see what improvements can be had there?

Edited March 1, 20224 yr by TurboKen

16 hours ago, TurboKen said:

It would be nice to see if anyone else could test this on there 3090 for comparison. I guess its safe to say I have a well assembled card. Which is good, because at its cost it will likely be in my machine for several more years. Sorry you didn't get the improvement we were all hoping to see Martin. Maybe you should attempt to remove the other side and see what improvements can be had there?

 

Somebody in this thread said they had the same issue.

Loads of people on the internet too. Miners were going frantic a few months ago and attaching heatsinks to the backplate, coolers on the backplate and all manner of stuff as their cards were throttling.

The issue has been on Tom's Hardware and all manner of YouTube tech guys have covered it. 

The issue first surfaced for people when they noticed that GPU temp was low but their cads were still throttling. It wasn't until Hardware Info was updated and GPUz that memory junction temp was available to view. Then it became clear what was happening. 

https://www.tomshardware.com/news/hwinfo64-adds-gddr6x-temp-monitoring-rtx30series

https://www.igorslab.de/en/gddr6x-am-limit-ueber-100-grad-bei-der-geforce-rtx-3080-fe-im-chip-gemessen-2/

Not sure what the variable is, why your card isn't affected. Might be better thermal pad quality, better pad installation. 

Not sure if I'll bother changing the pads on the GPU side or not.

What does make a huge difference is the "copper shim mod". Whereby the pads are replaced with copper shims. A thin layer of paste, a copper shim, and then more paste. 28 degrees scan be knocked of junction temp. Copper is electrically conductive of course, so care needs to be taken to protect surrounding components. 

 

Edited March 2, 20224 yr by martin-w

In case anyone is still interested in the DDR6X junction temp saga, I've just been reading a thread on the NVidia developers forum, where Linux users are going bonkers because NVidia aren't giving them to ability to monitor Junction Temp like we can in Windows.

The thread has been going on for over a year. Initially NVidia said they had a developer working on it, but then backtracked. Seems they are making excuses like "It's an alleged temp in HWiNFO. Which is clearly nonsense. We know that DDR6X throttles at 110 degrees and that's exactly what users are seeing when HMiNFO and GPUz hit 110 degrees. And of course, that's the only temp that isn't at low or moderate levels.

I'm suspecting that both NVidia and the third party manufacturers messed up and didn't anticipate the high temps of DDR6X and provide a sufficient cooler.

 

https://forums.developer.nvidia.com/t/request-gpu-memory-junction-temperature-via-nvidia-smi-or-nvml-api/168346

Edited March 9, 20224 yr by martin-w