What sort of memory temps are you getting fellow RTX 3090 owners.

Even at idle my memory is running at 64 degrees. 94 degrees in FS2020 (even just in menus) 

In COD Warzone 90 degrees. 

I'm aware that the memory chips can exceed 100 degrees before issues, but the manufacturers recommend a max temp of 95/110 degrees. 

The high memory temps are resulting in excessive fan noise, so I'm thinking of perhaps buying the AIO Hybrid kit for the card, or maybe even a custom loop.

What are your experiences guys? Same issue? Or did you find a solution? 

Edited February 8, 20224 yr by martin-w

In idle about 54°C, running FS2020 @4K Ultra with no frames cap also about 94°C .... 
Frames capped to 50 through NVCP about 90°C.

Noise is ok, but I am thinking about adding a top fan for may case to get temperatures down a bit.

 

I wanted to ask a question, since you are using eVGA based cards, are using the Precision X1 for monitor and adjusting over clocking and what resolution are you running on your monitor?

One thing I do know from my experience with the Titan X Pascal with high fan speeds is that if you are running hot, you may want to consider looking at the fan profile curves using Precision X1. The default fan curves are never optimum in your environment and you will need to adjusted it so that fan start running faster sooner before the temps rise to max on idle. You can save that profile and let it running every time you started it. 

The other issue is ambient temp inside the cabinet. 

You’re running a CPU with the fans blow heat inside the cabinet rather than a liquid cool solution taking heat to the radiator with the fans blowing the heat away from the cabinet. Ambient temperature in the room maybe not enough to bring down temps inside the cabinet while trying to exhaust heat out of the cabinet. You may want to try open the side of the cabinet while running flight sim, if you’re are experience overheating to stabilizes the temp under load. If that helps, then you going to much more aggressive cooling solution that you mention.

The power settings in Windows is also and issue. You have power setting on High Performance or Ultimate Performance. Technically your card is supposing to only overclock underload condition and down clock to its based clock when it’s not. If its clock speed is high all the time, it must mean the driver is not controlling it properly when you set your power settings in the nvidia driver. You may what to investigate different power settings, monitor the gpu clock speed and trying to sim to see if the clock stays high all the time or if it runs by some type adaptive state (meaning down clocked when not running under heavy load) when running the menus vs sim.

Even with Win 10, there is always some issue with that when try to running the GPU normally.

I will be able to say more once I get my system up and running.

 

 

1 hour ago, teletom said:

In idle about 54°C, running FS2020 @4K Ultra with no frames cap also about 94°C .... 
Frames capped to 50 through NVCP about 90°C.

Noise is ok, but I am thinking about adding a top fan for may case to get temperatures down a bit.

 

What does the memory temp drop to when capped at 50?

The top fan might make a slight difference. I notice if I open my side panel door in the EvolvX the mem temps drop from 94 to 90. I tried under clocking the card a bit and that only dropped the mem temp to 90 so minus 4. 

The EVGA AIO hybrid kit does drop it quite a bit so I understand, fan noise too. Non available at the moment though. I bet its the thermal pad change that helps too when the hybrid cooler is installed. 

2 minutes ago, martin-w said:

What does the memory temp drop to when capped at 50?

90°C

3 minutes ago, martin-w said:

The top fan might make a slight difference. I notice if I open my side panel door in the EvolvX the mem temps drop from 94 to 90

When I open the front door of the case I also notice a drop in temperatures, but only CPU, about 2°C less.
So hoping for the top fan to perform.

1 hour ago, BobFS88 said:

I wanted to ask a question, since you are using eVGA based cards, are using the Precision X1 for monitor and adjusting over clocking and what resolution are you running on your monitor?

 

I have EVGA Precision X1 installed yes. I'm not currently using it though, apart from experimenting with an underclock this morning. 

I have a 1440P monitor, so running at that native resolution. 

 

1 hour ago, BobFS88 said:

One thing I do know from my experience with the Titan X Pascal with high fan speeds is that if you are running hot, you may want to consider looking at the fan profile curves using Precision X1. The default fan curves are never optimum in your environment and you will need to adjusted it so that fan start running faster sooner before the temps rise to max on idle. You can save that profile and let it running every time you started it. 

 

Well yes Bob, thanks for that, but the fans are ramped up and its fan noise that's an issue too. Its only the memory temp that is high. GPU temp is lower. DDR6 seems to be running hot on the 30 series cards. Some have swapped thermal pads and dropped the temp of the memory a bit. 

 

1 hour ago, BobFS88 said:

You’re running a CPU with the fans blow heat inside the cabinet rather than a liquid cool solution taking heat to the radiator with the fans blowing the heat away from the cabinet. Ambient temperature in the room maybe not enough to bring down temps inside the cabinet while trying to exhaust heat out of the cabinet. You may want to try open the side of the cabinet while running flight sim, if you’re are experience overheating to stabilizes the temp under load. If that helps, then you going to much more aggressive cooling solution that you mention.

 

Yep, as I mentioned above, opening the side of the case drops the memory temp by just 4 degrees. Some people are resorting to directing a fan directly at the components on the top of the board apparently. 

 

1 hour ago, BobFS88 said:

The power settings in Windows is also and issue. You have power setting on High Performance or Ultimate Performance. Technically your card is supposing to only overclock underload condition and down clock to its based clock when it’s not. If its clock speed is high all the time, it must mean the driver is not controlling it properly when you set your power settings in the nvidia driver. You may what to investigate different power settings, monitor the gpu clock speed and trying to sim to see if the clock stays high all the time or if it runs by some type adaptive state (meaning down clocked when not running under heavy load) when running the menus vs sim.

 

Clock speeds aren't high all the time.

Windows power options are set to balanced. I recall power plans influence CPU not GPU. But anyway, both the CPU and GPU throttle back as they should. As I say, GPU temp is no issue, its low, its the DDR6 memory that runs hot. Memory Junction Temperature runs hot. Some are seeing over 110 degrees.

 

https://www.nvidia.com/en-us/geforce/forums/geforce-graphics-cards/5/430514/rtx3090-memory-junction-temperature/?topicPage=6

Edited February 8, 20224 yr by martin-w

9 minutes ago, teletom said:

90°C

When I open the front door of the case I also notice a drop in temperatures, but only CPU, about 2°C less.
So hoping for the top fan to perform.

 

Hmm, so pretty much minimal with a frame rate cap. From what I've seen you have to cap right down to 30 to make any difference.

Well yes, your top fan might help internal enclosure temps but I don't think you'll drop your card memory temp by much. 

I have an Asus 3090 with the same memory temperature limits, and I have a different approach to dealing with it that works well for me…

I do some casual mining on the side and I am controlling the mining rate by limiting the max GPU wattage to 205 watts.  On my card at 205 watts limit ((approx 52% TDP) the memory temperature stays below 94 degrees all the time with my fans running about 50%.  This has been a good balance for me between performance and heat.  The fans are still quiet at 50%.

I use my mining software to set the TDP/Wattage limit, but I believe you can achieve the same using something like Afterburner.   Disclaimer/Warning: at your own risk. Do your own research before trying.

J

 

Edit:  I do also set a 35 FPS limit inside P3D to limit excessive load on other components (like CPU).

Edited February 8, 20224 yr by jes0811
Add…

18 minutes ago, jes0811 said:

I have an Asus 3090 with the same memory temperature limits, and I have a different approach to dealing with it that works well for me…

I do some casual mining on the side and I am controlling the mining rate by limiting the max GPU wattage to 205 watts.  On my card at 205 watts limit ((approx 52% TDP) the memory temperature stays below 94 degrees all the time with my fans running about 50%.  This has been a good balance for me between performance and heat.  The fans are still quiet at 50%.

I use my mining software to set the TDP/Wattage limit, but I believe you can achieve the same using something like Afterburner.   Disclaimer/Warning: at your own risk. Do your own research before trying.

J

 

Edit:  I do also set a 35 FPS limit inside P3D to limit excessive load on other components (like CPU).

 

 

Thanks Jason. Useful info. 👍

Coincidentally with your reply... I just dived into Precision X1 and dropped the power target down to 80 with the temp linked. The result is that memory temp drops to 88 and lower fan noise. Much better.

I'll do some more experimentation tomorrow. I want to find out what sort of impact dropping the power target has on frame rate. I suspect that exactly what you did. 

 

 

 

Edited February 8, 20224 yr by martin-w

2 hours ago, martin-w said:

I have EVGA Precision X1 installed yes. I'm not currently using it though, apart from experimenting with an underclock this morning. 

I have a 1440P monitor, so running at that native resolution. 

Interesting, I will we going to that monitor as well. However I will starting off on the old monitor, which is a 1080p with my establish benchmarks first before moving up.

 

2 hours ago, martin-w said:

Its only the memory temp that is high. GPU temp is lower. DDR6 seems to be running hot on the 30 series cards. Some have swapped thermal pads and dropped the temp of the memory a bit. 

I wonder why that is, unless its was evga way of differentiating overclocking its memory to gain higher speeds by default. i could be wrong but something is driving it to reach those temperatures. Unless you needed to lower those power threshold in the first place, those memory temps look like they are on a runaway train. Something to watch for.

 

2 hours ago, martin-w said:

ep, as I mentioned above, opening the side of the case drops the memory temp by just 4 degrees. Some people are resorting to directing a fan directly at the components on the top of the board apparently. 

If its was me I would be going for outside lol. I do that to old old system all the time. I open my window where it sits by.  But extra cooling can help a great deal.

 

2 hours ago, martin-w said:

But anyway, both the CPU and GPU throttle back as they should.

Good deal on that, at lease that not an issue.

 

It seems like you found a solution is lowing the power limits has help some with the temps which I think you can just achieve using the Precision X1

35 minutes ago, BobFS88 said:

I wonder why that is, unless its was evga way of differentiating overclocking its memory to gain higher speeds by default. i could be wrong but something is driving it to reach those temperatures. Unless you needed to lower those power threshold in the first place, those memory temps look like they are on a runaway train. Something to watch for.

 

Its not just EVGA cards. Other manufactures cards do the same. It seems to be the nature of DDR6.  Although as I mentioned, quality of manufacturers thermal pad installation seems to be a variable.

Also worth remembering that the 3090 FTW 3 Gaming Ultra is a manufacturer overclocked card.

In terms of power limits, I just did some testing in Farcry 6.

Linking power limit with temp, so 80% power limit and 78C temp limit: 128 frames per second average.

Unlinking power limit and temp, so 80% power limit and 83C temp limit: 128 frames per second.

Not reducing power limit so stock settings: 132 frames per second.

So only about a 3% loss in performance. And that's just a rough estimate. Probably within the margin or error. 

 

 

 

 

 

 

Edited February 8, 20224 yr by martin-w

In FS2020, on Ultra setting. Didn't seem to impact frame rate at all by reducing power limit to 80%. Still high 70's to high 90's  in terms of frame rate. 

6 minutes ago, martin-w said:

In FS2020, on Ultra setting. Didn't seem to impact frame rate at all by reducing power limit to 80%. Still high 70's to high 90's  in terms of frame rate. 

I think in the high 70's is where you would like to be IMO and lower if you can. The lower the better but that may come later with better cooling.

2 hours ago, BobFS88 said:

I think in the high 70's is where you would like to be IMO and lower if you can. The lower the better but that may come later with better cooling.

 

 

Not for 3090 memory temp. Mid 80's is a good temp for 3090 memory and safe. 3090 memory idles at 50 to 60. 80's isn't that high for GPU temp either really these days. 

Of course what we deem acceptable is often subjective rather than based on technical facts. But yes, Micron and Samsung confirm that 95 or less is in the normal range for DDR6. Above 110 potential damage is feasible if sustained long term.

I think we would be doing well to get 3090 memory down to mid 70's probably requiring a custom loop. Silicon lottery applies though, there's bound to be some examples that can do it.

Edited February 8, 20224 yr by martin-w

Nice info. I may experiment. My 3090 sounds like a ram air turbine at times.