News Replacing GeForce RTX 3090 Thermal Pads Improves GDDR6X Temps By 25C

[Admin]

YouTuber demostrates the benefits of swapping the GeForce RTX 3090's factory thermal pads for aftermarket ones.

Replacing GeForce RTX 3090 Thermal Pads Improves GDDR6X Temps By 25C : Read more

 

[husker]

I suspect this does not come as a surprise to Nvidia. Could be they saw this as a low cost way to discourage miners while keeping game benchmarks high.

 

[King_V]

I don't know. Seems sloppy on Nvidia's part. I don't think they were planning on a mining explosion when their cards released - particularly during design/manufacture.

Seems like just a way to pad profits slightly. And a real shame.

 

[btmedic04]

I suspect this does not come as a surprise to Nvidia. Could be they saw this as a low cost way to discourage miners while keeping game benchmarks high.

I don't know. Seems sloppy on Nvidia's part. I don't think they were planning on a mining explosion when their cards released - particularly during design/manufacture.

Seems like just a way to pad profits slightly. And a real shame.

Built in obsolescence and pinching pennies to increase profit margins. If we start to see a lot of 3090 FE failures in 3 years, then i'd say its safe to assume that was nvidias plan all along

 

[sharkbaithohaha002]

Would this void my warranty though?

 

[Quarkzquarkz]

That's terrible tweaking. I used thermalright odyssey 1.5mm thermal pads on my 3090 founders edition backplate and getting 117mh/s continuous with 46c gpu temps and 65% fan speed with a power limit at a cool 280 watts across. Also added a heatsink and I'm running 20.04 ubuntu headless. Tested with trex, nbminer, gminer, phoenixminer, and ended up switching between t-rex and gminer.

Whoever wrote that article doesn't realize they can probably push it to 120mh/s all day without worry.

 

[Sethiatspace]

That's terrible tweaking. I used thermalright odyssey 1.5mm thermal pads on my 3090 founders edition backplate and getting 117mh/s continuous with 46c gpu temps and 65% fan speed with a power limit at a cool 280 watts across. Also added a heatsink and I'm running 20.04 ubuntu headless. Tested with trex, nbminer, gminer, phoenixminer, and ended up switching between t-rex and gminer.

Whoever wrote that article doesn't realize they can probably push it to 120mh/s all day without worry.

Not all cards are created equal, some 3090 FE (or not) have much better results than others. A lot of cards don't reach 120 MH/s or they will have incorrect/rejected shares or instability, which are both equally bad.
Also, gpu temp is not a metric you should look at, it means nothing in the case of mining as it's the memory that's hit hard. You can have 50C gpu temp and 110 memory, check with Hwinfo.

 

[Olle P]

... Seems sloppy on Nvidia's part. ...
Seems like just a way to pad profits slightly. And a real shame.

Given that the heatsink is crazy expensive (manufacturing cost) one should expect Nvidia to cut all other available corners to reduce their loss on each sold card.

 

[Quarkzquarkz]

Not all cards are created equal, some 3090 FE (or not) have much better results than others. A lot of cards don't reach 120 MH/s or they will have incorrect/rejected shares or instability, which are both equally bad.
Also, gpu temp is not a metric you should look at, it means nothing in the case of mining as it's the memory that's hit hard. You can have 50C gpu temp and 110 memory, check with Hwinfo.

Unfortunately, for linux users theres no way to check for vram temps outside hiveos and windows 10. But rest assured I've been running my 3090s like this for a little over 2 months now and all have thermal pads replaced, heatsinks added, low temps, and proper cooling and all happily at 117mh/s

 

[King_V]

Given that the heatsink is crazy expensive (manufacturing cost) one should expect Nvidia to cut all other available corners to reduce their loss on each sold card.

Wait, did I miss something somewhere? Nvidia's losing money on each card sold?

 

[Sethiatspace]

Wait, did I miss something somewhere? Nvidia's losing money on each card sold?

There were rumors that the FE cooling solutions are too expensive and are costing nvidia real money.

 

[spongiemaster]

There were rumors that the FE cooling solutions are too expensive and are costing nvidia real money.

They're not losing money on a $1500 3090. If they were, they'd be losing $100's on a 3080 which uses the same basic design but is a bit smaller. Nvidia isn't losing money on any of their FE models. Let's not kid ourselves.

 

[VforV]

There were rumors that the FE cooling solutions are too expensive and are costing nvidia real money.

😆🤣
That's all I have to say to that rumor and who ever believes it.

 

[JarredWaltonGPU]

I don't know. Seems sloppy on Nvidia's part. I don't think they were planning on a mining explosion when their cards released - particularly during design/manufacture.

Seems like just a way to pad profits slightly. And a real shame.

Having tested and reviewed previous Founders Edition cards, I sometimes wonder if Nvidia is just trying to play it a little safe so as not to irritate its AIB partners. All of the AIBs have had better cooling solutions for YEARS. I was actually really hopeful with the 30-series FE cards (before testing), because it looked like they would be better at cooling.

With cards in hand, however, while they're quiet and run well for gaming, the backs of the cards still get very hot on the 3080 and 3090. Maybe not as hot as the 2080 Ti and 2080 Super, but close enough. This should have absolutely been addressed with the 3090, since Nvidia knew it would have GDDR6X on both sides of the PCB and that the back would get much hotter. I think it ended up going for a "cool" look -- which I actually do like -- over better cooling.

Fundamentally, it's very simple to put on better thermal pads and have more contact between the memory and heatsink. Nvidia should have done this on the 3090 and 3080. Even without mining, the VRAM temps are higher than normal and dealing with that rather than reducing the memory clocks would have been the right approach. It's obvious from the way the FE cards behave (GPU temps stay low, VRAM hits 110C, fan speeds go from 40% to 100% in an attempt to keep the memory cool) that the engineers recognized what was happening and built in a "failsafe" of sorts in the firmware. Possibly it was too late for the first batch of cards.

Hopefully the future updates (3080 Ti or whatever) tweak the design just a bit. We're talking about pads that probably would cost Nvidia $1 per card extra. On a $300 card, that might be a problem, but on a $700 or $1500 card? It's nothing.