How Does HWInfo Know It's Thermal Throttling But No GPU Temp Suggests That?

[ThisIsMyAcct]

According to HWInfo my GPU temp can be as low as 41°C, but the "Power Limitation - Thermal" reports Yes. How does it know that?

This must imply that there is another thermal sensor on the card, but nowhere else does it show any other GPU temps. I can see the clocks slowing down which also suggests thermal throttling, but I was curious how it's thermal throttling if there's no temperature read out on HWInfo.

I guess its completely possible that some sensors are not set up to give temperature read outs to the computer but rather just internally with the GPU. Would HWInfo be able to extract this information one day?

My card is the MSI RTX 3080 Gaming X Trio.

 

[Martin]

The NVIDIA driver is responsible for providing this information, so there are 2 possible explanations of this behavior:
- either this flag is not correct and NVIDIA needs to fix this in the driver, or
- it's correct and (as you assume) some other component is reaching the thermal limit
In the second case it would most likely be the GPU memory, which can reach quite high values, unfortunately no one (except NVIDIA) is currently able to determine the actual GPU memory temperature:

RTX 3080/3090 Memory/VRM temperature sensor

Can this be added in a future hwinfo? The board seems to have a way to read this, since people have been getting "Thermal" Limit even with a 55C CPU temperature. https://www.overclock.net/threads/official-nvidia-rtx-3090-owners-club.1753930/page-218#post-28667624 Igor's website seems to show...

www.hwinfo.com

We have spent some effort how to determine this, but haven't been able to figure it out yet. Still keep trying...

 

[ThisIsMyAcct]

The NVIDIA driver is responsible for providing this information, so there are 2 possible explanations of this behavior:
- either this flag is not correct and NVIDIA needs to fix this in the driver, or
- it's correct and (as you assume) some other component is reaching the thermal limit
In the second case it would most likely be the GPU memory, which can reach quite high values, unfortunately no one (except NVIDIA) is currently able to determine the actual GPU memory temperature:

RTX 3080/3090 Memory/VRM temperature sensor

Can this be added in a future hwinfo? The board seems to have a way to read this, since people have been getting "Thermal" Limit even with a 55C CPU temperature. https://www.overclock.net/threads/official-nvidia-rtx-3090-owners-club.1753930/page-218#post-28667624 Igor's website seems to show...

www.hwinfo.com

We have spent some effort how to determine this, but haven't been able to figure it out yet. Still keep trying...

Thank you for explaining so clearly and for even taking your time to share your knowledge. It's greatly appreciated and I love your team's work!

Edit: I want to mention that EVGA's card seems to provide the temperature read outs for using their EVGA software. I don't know if that will help you identify how to access these thermal readings but hopefully it does.

 

[Martin]

EVGA is a different case - they use additional sensors placed close to the GPU memory chips (so it's the external GPU memory temperature) and it works on their cards only. This is already supported in HWiNFO.
But GDDR6/6X have native internal temperature sensors reading the junction temperature, which is substantially higher than what can be measured externally. This junction temperature caused some confusion during Navi10 release as people became frightened after knowing it. It was clarified by AMD that this is the internal value, which also has a higher limit.
But NVIDIA has chosen to keep this secret and doesn't want people to know it. Probably to avoid a similar panic like in the Navi10 case, or they have some other reasons...

 

[Rex Seattle]

I have a ASUS ROG Strix 2080Ti which functions very well, yet one of the values in this series is always 'yes'. Sometimes more than one. The card is overclocked to about 2Ghz using Nvidia's own OC scanner.

I used to worry about it, but I've had yet to see it do anything unexpected, and the temps have never exceeded 84C

 

[Martin]

This "Utilization" reason mean the GPU is not under sufficient load to drive the clocks higher. So it's not issue at all

 

[Rex Seattle]

Awesome, thanks! Since that's usually the red one, I feel much better now!

 

[ThisIsMyAcct]

So I did a tear down to replace all the TIM on my MSI RTX 3080 Gaming X Trio - turns out one of the thermal pads was folded over on itself!
That's probably the reason why I was getting thermal throttles at 45°C.

I used Gelid GP-Ultimate on the VRAM, Gelid GP-Extremes on the back of the VRAM. I also used Prolimatech PK-3 for the VRAM plate to the heatsink plate and the GPU die.
I was able to overclock it with the MSI RTX 3080 Suprim X VBIOS with +220 clock and +1500 mem and got the top 2% GPU score (19472) on 3DMark Time Spy → https://www.3dmark.com/spy/17783288
Using the stock Gaming X Trio VBIOS I maxed out at 19138 → https://www.3dmark.com/spy/17785586

Thanks for the awesome update HWInfo to include the GDDR6X junction temps! I wish I had this information prior to modding to compare temperatures. I look forward to seeing the GPU die junction temps!

 

[ThisIsMyAcct]

Since ethereum mining is the most thermally stressful thing I have, I wanted to share the results of mining over 6 hours post-mods.
I would love to see what people with a stock MSI RTX 3080 Gaming X Trio are getting.

This is VBIOS flashed to MSI RTX 3080 Suprim X and overclocked.