Hot News? More Like... HWInfo64 Now Shows Hot VRAM for RTX 3080/3090!

[Rym]

It's kinda bittersweet news.

 

This is great news because if memory goes pop and gets permanently damaged then miners won't leech our GPUs which are meant for gaming and not mining, and bonus prices will drop. Wtb miner 3080 100$ must be working as intended with proof

 

But it's bad news because this throttling lowers gaming performance and lowers the lifespan of the GPU.

[HenrySalayne]

On 1/28/2021 at 8:24 AM, DaJakerBoss said:

provided AIB's with the information to account for those temperatures in cooling.

Haven't you learned anything from the MLCC debacle? AiBs get neither adequate informations nor the time to test it.

[Orangeator]

9 hours ago, linuxChips2600 said:

Could you post some screenshots? And what were the tests that you ran (i.e. was it memory-intensive workloads)? Also it seems that your TUF card is air-cooled and not liquid-cooled?

Name the tests you would like for me to run and I will share screenshots of the results. Honestly, I believe the reason there is such a difference in temperature is because the heatsink that is cooling the memory is completely detached from the major heatsink that is cooling the GPU die. It's also likely the reason why the TUF is the coolest running AIB card (as far as the GPU die is concerned).

[Moonzy]

1 hour ago, Orangeator said:

Name the tests you would like for me to run and I will share screenshots of the results. Honestly, I believe the reason there is such a difference in temperature is because the heatsink that is cooling the memory is completely detached from the major heatsink that is cooling the GPU die. It's also likely the reason why the TUF is the coolest running AIB card (as far as the GPU die is concerned).

For VRAM stress test, probably mining ethereum is a decent enough stress test

 

I'm not surprised if 3080 is better cooled than 3090 because there's no backside VRAM so they get more area and less dense heat per surface area

[Orangeator]

3 minutes ago, Moonzy said:

For VRAM stress test, probably mining ethereum is a decent enough stress test

 

I'm not surprised if 3080 is better cooled than 3090 because there's no backside VRAM so they get more area and less dense heat per surface area

I don't have any of that set up, you need to attach a guide for that lol.

 

Yes, that likely is contributing to this as well.

[RejZoR]

I have to check this at what temp mine run, but my Palit RTX 3080 GamingPro cools the VRAMs from both sides. On one side it's alloy frame covering VRAM and all the VRM components and the metal backplate also has thermal pads on the back of VRAM chips too. And let me tell you, the backplate gets so hot during gaming it's literally not possible to hold it even for 2 seconds. It's that hot. Which should mean the backplate is actually pulling heat from somewhere. And it's just the contact of back of the VRAM chips. So I'm really interested in what temps I'll have...

[linuxChips2600]

4 hours ago, Orangeator said:

Name the tests you would like for me to run and I will share screenshots of the results. Honestly, I believe the reason there is such a difference in temperature is because the heatsink that is cooling the memory is completely detached from the major heatsink that is cooling the GPU die. It's also likely the reason why the TUF is the coolest running AIB card (as far as the GPU die is concerned).

Per this (note that I jumped to the relevant spot already):

Try video rendering/export or any content creation task (e.g. blender "Victor" benchmark) that basically maximizes out workloads which rely on a ton of VRAM usage/activity.  I won't be surprised if the temps peak up to at least 90 C within the first 5 - 10 min.

[RejZoR]

Ok, after playing Dead Space 3, but with RTGI ray tracing added, it says VRAM is at 92°C. I'll have to try something even more demanding...

[linuxChips2600]

1 minute ago, RejZoR said:

Ok, after playing Dead Space 3, but with RTGI ray tracing added, it says VRAM is at 92°C. I'll have to try something even more demanding...

Have you tried DLSS enabled (if applicable)?

[Orangeator]

58 minutes ago, linuxChips2600 said:

Per this (note that I jumped to the relevant spot already):

Try video rendering/export or any content creation task (e.g. blender "Victor" benchmark) that basically maximizes out workloads which rely on a ton of VRAM usage/activity.  I won't be surprised if the temps peak up to at least 90 C within the first 5 - 10 min.

After rendering the Victor optix (Raytracing version) benchmark two times back to back these were my results on HWinfo64.

[linuxChips2600]

17 minutes ago, Orangeator said:

After rendering the Victor optix (Raytracing version) benchmark two times back to back these were my results on HWinfo64.

Should we start a separate thread where we compile a list of various RTX 3080/3090s' which do NOT thermal throttle even while maxing out VRAM usage?

Also I found two different Ethereum mining clients which you could give a shot just to see if you can make it hit 80+C (this website is what prompted me to recommend these particular ones over any others; hopefully @RejZoR at least one of these clients will suffice for maxing out your VRAM too):
https://github.com/nanopool/Claymore-Dual-Miner
https://github.com/openethereum/parity-ethereum

Finally, did you get any idea of how hot the back-plate got during the Victor benchmark?  If you don't have some non-reflective tape and a thermal camera, try very carefully using the backs of your fingers to get a rough idea of how hot it gets (a trick I learned from my father who's very good at cooking); that way you are able to pull your fingers back as quickly as possible in case if the back-plate is actually hot (which is very likely imo).

[Orangeator]

1 hour ago, linuxChips2600 said:

Should we start a separate thread where we compile a list of various RTX 3080/3090s' which do NOT thermal throttle even while maxing out VRAM usage?

If you start the thread I will contribute what I can haha.

1 hour ago, linuxChips2600 said:

Also I found two different Ethereum mining clients which you could give a shot just to see if you can make it hit 80+C (this website is what prompted me to recommend these particular ones over any others; hopefully @RejZoR at least one of these clients will suffice for maxing out your VRAM too):
https://github.com/nanopool/Claymore-Dual-Miner
https://github.com/openethereum/parity-ethereum

Both of those require a lot of setup which I neither have the time nor patience to deal with, sorry lol. There are config files and I have to go and figure out what the different commands are, that's too much. I have never set a miner up before. 

1 hour ago, linuxChips2600 said:

Finally, did you get any idea of how hot the back-plate got during the Victor benchmark?  If you don't have some non-reflective tape and a thermal camera, try very carefully using the backs of your fingers to get a rough idea of how hot it gets (a trick I learned from my father who's very good at cooking); that way you are able to pull your fingers back as quickly as possible in case if the back-plate is actually hot (which is very likely imo).

I did touch the backplate, it gets hot, but I can hold my fingers on there for 5-8 seconds before I have to pull them off out of fear of being burned lol.

[Craftyawesome]

4 hours ago, Orangeator said:

Both of those require a lot of setup which I neither have the time nor patience to deal with, sorry lol. There are config files and I have to go and figure out what the different commands are, that's too much. I have never set a miner up before.

Try

t-rex -a ethash -B --benchmark-epoch 391

with https://github.com/trexminer/T-Rex/releases/tag/0.19.9

 

It may also be worth trying with the base suggestion from whattomine.com (+0 core, +500 memory, 60% power target)

[Fnige]

On 1/28/2021 at 5:39 PM, leeyz717 said:

Must be why EKWB released these babies:

 

https://www.ekwb.com/shop/ek-quantum-vector-fe-rtx-3090-d-rgb-silver-special-edition

 

Quite an engineering marvel really. Active backplate cooling to chill those memory modules on the back of the PCB. Though $330 would set you back more, considering you just spent that hefty $1800-2000 of yours on that RTX3090 .

 

 

... Why is it this expensive?

[Duskie]

On 1/28/2021 at 6:39 AM, leeyz717 said:

Must be why EKWB released these babies:

 

https://www.ekwb.com/shop/ek-quantum-vector-fe-rtx-3090-d-rgb-silver-special-edition

 

Quite an engineering marvel really. Active backplate cooling to chill those memory modules on the back of the PCB. Though $330 would set you back more, considering you just spent that hefty $1800-2000 of yours on that RTX3090 .

 

 

Unfortunately the backplate on those isn't actually "active", notice how EK is very careful in their wording. It's very likely there's just a thermal pad connecting the coldplate of the block to the backplate. AquaComputer has active backplates with their Kryographics NEXT stuff though.

On 1/28/2021 at 9:03 AM, linuxChips2600 said:

Does anyone have water-cooled VRAM (note VRAM not VRM) temps to post for reference (Google mixes the two terms up with its "fuzzy search" functionality; no luck with using the term "memory" instead in Google search nor searching this forum directly either)?  And yes you must post the temps under sustained 100% GDDR6X load when thermal equilibrium has been (basically) reached; otherwise it doesn't count.

I have a 3090 FE with a Bitspower block on it, the maximum GDDR6X junction temp i've seen is 78c with a loop of Metro Exodus on maximum settings. The front-side modules are cooled by the cold plate so I assume they're at 50-60c, but the ones on the back sit around 70-80c due to it just being a passive backplate. The surface temperature of the backplate is around 40-50c under 400w overclocked load.

 

Something also to note is that the backplates that come with FE waterblocks are connected to the back of the core via thermal pads whilst the stock backplate is not. This helps to reduce heat soak in the PCB and does help core temperature by a degree or so whilst reducing the total amount of hotspot around the core. I imagine this isn't included on the stock backplate because the additional heat from the back of the core would push the VRAM to the point of overheating/throttling, which is what we're seeing with mining on 3090 FE's.

[Orangeator]

12 hours ago, Craftyawesome said:

Try

t-rex -a ethash -B --benchmark-epoch 391

with https://github.com/trexminer/T-Rex/releases/tag/0.19.9

 

It may also be worth trying with the base suggestion from whattomine.com (+0 core, +500 memory, 60% power target)

Thanks! Running the Eth miner in that folder at full tilt had my memory hit 92C! So yeah, 18 degrees lower compared to what this article is claiming, but super toasty! That said, no other load I have tried could get it above 72c... So a non-issue for me haha. 

[LucisZ]

On 1/28/2021 at 1:03 AM, linuxChips2600 said:

Does anyone have water-cooled VRAM (note VRAM not VRM) temps to post for reference (Google mixes the two terms up with its "fuzzy search" functionality; no luck with using the term "memory" instead in Google search nor searching this forum directly either)?  And yes you must post the temps under sustained 100% GDDR6X load when thermal equilibrium has been (basically) reached; otherwise it doesn't count.

 

I have a 3080 with the FE block, mining at +1500 VRAM and the memory is still only around 66C (-500 on VRAM it will drop down to like 58).  I think this probably as memory intensive as you can get.  Notice my water is 30 degree. Since the cooling for component is completely depended on that.  But even at 40C water I'd imagine that the memory won't get over 76C as it's pretty linear.  

[linuxChips2600]

5 hours ago, Orangeator said:

Thanks! Running the Eth miner in that folder at full tilt had my memory hit 92C!

How long did it take to reach those temps (e.g. ~10 min, ~20 min)?

[Orangeator]

1 hour ago, linuxChips2600 said:

How long did it take to reach those temps (e.g. ~10 min, ~20 min)?

Frankly, it hit 92 in around 5 minutes, I continued to run it for another 20 but it never budged... 

[leeyz717]

On 1/30/2021 at 6:56 PM, Duskie said:

Unfortunately the backplate on those isn't actually "active", notice how EK is very careful in their wording. It's very likely there's just a thermal pad connecting the coldplate of the block to the backplate. AquaComputer has active backplates with their Kryographics NEXT stuff though.

Haha, sorry about that. They did say that it uses a hybrid design, the provided pics does suggest that you might be right . Regardless, the need for such a product to be made just shows how hot those memory modules are.

 

I recently just scored a Zotac RTX 3080 Holoback, and I'm frankly disappointed with the aftermarket cooler.  Just got a EK waterblock to cool it down better, so that should tame the temps better. Was going to complain about the "premium" pricing, but those scalpers really does change perspectives.  

[noovator]

Hi all,

 

I have two rtx 3090. One is Gainward Phantom, positioned just below cpu air cooler. The other is Zotac Trinity.

 

When mining Gainward 3090 reaches stable 95°C.

 

Zotac goes all up to 106 - 108°C when mining.

Undervolting, underclocking doesn't help. 10cm fan on the top of the 3090 Trinity doesn't help neither.

 

 

[Massey_Ferguson]

I've got ASUS TUF 3080.

 

ETH mining through NiceHash, i used first their default settings for ~48 hours straight:

Power: 68

Core clock: -150

Memory clock: +900

Fan auto.

I was running 104-106 C. The card did not thermal throttle.

 

~94 MH/s

 

Once I learned about the memory heat problem, I've used:

Power 63

Core -300

Memory clock: 0

Fan 85 %.

 

I run stable at 88 C. with ~82 MH/s.

 

Do you think the latter is sustainable, or is it still too hot to run long term?

 

Playing R6 Siege with no OC and maxed out settings gives memory temps of 82 C. 

 

[Massey_Ferguson]

22 minutes ago, Nacht said:

 

Graphics cards aren't designed to mine on 24/7 try gaming on it.

I play on it every day, but would not say no to 160 usd/month mining profit. 

[linuxChips2600]

3 hours ago, Massey_Ferguson said:

Do you think the latter is sustainable, or is it still too hot to run long term?

tbh I'm not comfortable running any of my electronics above 85 C for an extended period of time.  But I'd say that given your results posted you should not even consider running at power targets any higher than ~65%.

Keep in mind that as stated in my OP the minimum max operating temperature of GDDR6X memory chips is 95 C, so although Tjunction may be set at 110 C by Nvidia drivers for now, I would strongly caution against running the memory on an RTX 3080/3090 at 95 C or higher for any period of time, as once the transistors fail it's game over unless if you are tech and business savvy enough to find replacement GDDR6X memory chips and do the actual replacement yourself...

Wonder if the next big shortage will become GDDR6X memory chips, especially if Nvidia still has the guts at this point to release the RTX 3080 TI (or hell even the 3080 20 GB version) within the next year or so...  Does anyone know for sure that GDDR6X is a Micron-exclusive product (since it appears to be the case at least to me just from Micron's website)?

[Maticks]

Maybe a really good reason to have a full cover waterblock this gen.