3090 FE RMA opinion?

[tintin82]

Hi Folks,

I was lucky enough to get my hands on a 3090 FE for MSRP.

Initially I was really happy as coming from a 1080ti is was a big jump in performance. But I noticed in some RTX titles (Quake RTX) that the card quickly became akin to a leaf blower with the fans ramping up to 100%. A quick Google search revealed that this was probably due to the temperature of the VRAM.

Many recommended replacing the thermal pads which I am not keen to undertake. However I found that by exerting upward force on the cooling shroud the temperatures and fan speed dramatically reduce. It would appear that the cooler and VRAM are not making enough contact without the upward force.

I have now made a Lego tower to support the GPU and keep the cooler in contact with the VRAM and now the fan rarely goes above 60 - 70% which is much more in line with expectations.

However to have paid £1,400 for a GPU that needs a Lego prop is not ideal. I asked the retailer if this was acceptable and they didn't comment, but rather started the RMA process. I certainly don't want to go and send this card back if this is normal, and then at best get a replacement that does exactly the same thing, or at worst be stuck with no card for however long.

The card is great, and with my Lego mod it is nice and quiet, but my OCD has me annoyed that I may have a duff card. What is your opinion. Keep it, or send it back?

 

[tintin82]

Most likely GPU sag or he's forcing the thermal pads to touch the gddr6x as they have poor positioning or contact.

Personally I'd leave it until supply is no longer an issue. Chances are that you will be waiting weeks for a replacement.

Do you have another card to use in the mean time? If not wait 6 months then RMA it. It's not as if you cannot use it currently as is.

Yeah this is where I am at. Pretty sure it's gpu sag, and although I have a 1080ti, it took me long enough to get the 3090 so I would rather enjoy it than deal with an open ended rma process.

 

[tintin82]

Have you tried laying the PC on it's side? This will take the stress off the slot and card. The GPU will sit nice and square, with no external stress.

You need to determine if your lego is just helping the GPU sit level or if it's using gravity/tension/ pressure to apply contact force due to poor internal assembly. Ie a faulty GPU.

The 3090fe is a big heavy card. I've got a stand supporting my 3090fe. (Cheap from the rain forest)

Even though it's secured with the backplate screws, and the motherboard slot is strengthened, I still use a support. I set it to the correct level with the PC case laid on its side, before turning it upright.

Without a doubt it sags without the support which in turn leads to the fan running at 100%. With the support in place the improvement is instantaneous and the fan reduces to 60%. The positioning of the support does result in different temp drops. If positioned under the Vram it will drop more than if at the end of the card, but both lead to very good improvements. Is this a RMA worthy reason though, or likely to receive another card with exactly the same issue.

 

[tintin82]

So would I be right in saying, the case/ motherboard slot configuration is not fully supporting the weight of the 3090 and causing it to sag/ flex under its own weight?

If so, a warranty replacement isn't going to fix that, you just need adequate support which you've done.

There is a thread on here, with lots of people using all sorts of methods to combat gpu sag. Your not the first person to use Lego

I think it is the sheer weight making it sag.

I have heard of the sag issue, but more from a cosmetic POV, never a situation where it actually affects performance as in this case, hence my concern

 

[tintin82]

I wouldn't go replacing anything or taking the card apart as you risk voiding warranty. Simply live with the lego for now (paint it black or cover the lego with black tape if it triggers your OCD) then RMA it in 3-6 months when things have calmed down.

I think this is the way. It may be possible that there is nothing wrong with the card and the lego mod simply improves the cooling (I found someone else on YouTube recreate this problem). I applied a underclock of 0.85 and locked the core at 1750mhz. Card was completely silent with zero coil whine for hrs in BF5, fan did not go over 40%. Really is an amazing card.

 

[tintin82]

Hi all.

Thanks for all you input and suggestions. I have conducted some tests with quite remarkable results. All tests consisted of running Quake RTX for 10 mins with all stock settings and a PL of 114% of maximum heat.

I tried custom fan curves and underclocks, but this was more a remedy for the symptoms, not a cure. I also picked up some heatsinks and a small 60mm fan for the VRAM (all for under £10). This knocked a few C off, but nothing extraordinary.

However yesterday I noticed that my original stand still had a very slight amount of give, so I added some of the smooth flat pieces of Lego to make it bar tight. Eliminating this last amount of give resulted in a further 10c drop! Unreal! It is very tight, but the tension is on where the GPU screws into the case, not the PCIE slot.

Case upright, no stand - Fans 100%, VRAM 112c, throttling evident
Case on side, no stand - Fans 65%, VRAM 108c

Case upright, mk1 stand - Fans 55%, VRAM 102c
Case upright mk2 stand - Fans 45% VRAM, 96c
Case upright mk2 stand plus heat sinks - Fans 45%, VRAM 92C


Based on purely what I have observed I would guess that many who are replacing thermals pads may not actually need to. The pads may not be the issue, more the contact.

The easy way to check yourself is to run a game and observe HWinfo. Wait until the memory temp is steady and then simple pull the GPU upward gently. The results in my case were instant.

I am very happy that I managed to drop the temps by 20c and reduce the fan speed by half without opening the GPU. I doubt changing pads will do much better. Regards the RMA.... I think the likelihood of having the exact problem with a replacement is high as so many are replacing pads, so I may keep it.


If anyone else has a similar experience it would be great to hear your results.

 

[tintin82]

In all honesty that card isn`t right and you shouldn`t have to be adding makeshift stands to get it working correctly. I`ve been following this thread since it was started and although some people are gretting high temps this is the first time I`ve heard a case like this.
My 3080FE was quieter than my 1080ti before I added the water block and the fans never went to 100%, the chances of you getting a card back with the same problem are very slim, so I would RMA it but it`s your call at the end of the day.

Yeah this is where I am conflicted. The reason that I even went down this rabbit hole was this video . So it is not an isolated incident, just perhaps a less known fix.

 

[tintin82]

Interesting, based on your findings here is what I think.

When you actually get your hands on the OEM white pads you will understand why there is air space being created. They are so soft and have no retention in them to return to their original state (or even try to). So by you flexing the card, you are creating the air space between the metal and pad. I think once you flex it once, you are screwed, the air space is created and you will need to open it up to reshape the OEM white pad.

These OEM white pads feel like marzipan, so soft.

I think the aftermarket pads act like sponges and will give you that consistent contact between the pad and metal.

I am happy my pad change and I can flex my card (3080) with no variation in temps. (very consistent).

That sounds like a very reasonable explanation! The temperature changes as I insert and remove the lego tower is instant, indicating an air gap as the contact is lost.

 

[tintin82]

I’ve just come across this thread and I’m pretty much in the same position as OP. I found that applying pressure underneath the card drops the VRAM temps by at least 15c.

I haven’t replaced any pads, just put these two braces underneath.

The braces are applying a bit of pressure so I’m a little bit concerned it might damage the card, not sure..

I’m convinced this is the case with every 3090 FE as a friend of mine with a 3090 FE is doing the same thing.

Great to hear you had a similar success. Its great to drop those temps without taking the card apart. My first attempt dropped them temps by a fair amount, then I took it one step further and applied a load of pressure and the temps went through the floor. I fairly am confident this will not damage the card as the resistance is taken by where the card screws into the back of the pc. My lego stand is bar tight against the cooler. Works a treat.

 

[tintin82]

The GPU brace support off the bay (as suggested by Mushta earlier in this thread) arrived today so I've set it up under my 3090 with a very conservative amount of pressure. GPU stress testing wuing Quake II RTX demo (the best stress test) - doing this has not lowered Vram temps or had an adverse effect on core temp (ie no change for either).......... so......

1. Either I'm not applying enough pressure with it?
2. It's not in the right location on the card?
3. A combination of 1 and 2?
4. I don't need to do it anyway for my card?
5. It's all ********?

Thoughts?

I would say 4. Your first post on this thread says in Quake RTX the VRAM doesn't go past 98, with fans at 60-70 %. Seems pretty good so I don't really know what you are trying to achieve?

 

[tintin82]

Possibly, I hope it's #4 anyway.

What I'm trying to achieve? Lower Vram temps obviously, with a potentially very easy fix that doesn't involve opening up the card and risking warranty.

I should point out that 98 degrees is when running my most conservative undervolt on the card which is 0.8v at 1725 clock speeds. If I run the card at stock for example the Vram temps are obviously higher.
The Quake 2 RTX demo is an extreme case though and does not represent normal gaming, even other RTX games, so most of the time VRam temps aren't really a problem for my card especially as I always run undervolted now. But, I always like lower temps so a cheap, quick and 'non invasive' solution was worth a shot....

Have you tried a small fan and heatsink? This wipes another 4-5 C from my vram temp. Cost me under £10.