How to display MEM temps on Hive OS

When mining ETH on Hive OS, it’s very useful to know your CORE and MEM temps. Understanding these temperatures might prevent long-term damage to very expensive GPUS or you might use MEM temp to understand if your GPUs have poorly installed thermal pads.

CORE temps have long been visible and as of April 2020 you can now see MEM temps for NVIDIA’s cards that have GDDR6 and GDDR6X memory.

These NVIAIA cards are supported: RTX 3090, RTX 3080 Ti, RTX 3080, RTX 3070 Ti, RTX 3060 Ti, RTX 2080 Ti (GDDR6, limited), RTX 2080 (GDDR6, limited), RTX A2000 (GDDR6), RTX A4000 (GDDR6), RTX A4500 (GDDR6), RTX A5000 (GDDR6)

There are two methods. the first one is the easiest and should just work. For one of my rigs, I had to do the second.

1. Upgrade to firmware 0.6-214@220331

Simply upgrade the firmware on each rig to the latest version. More specifically anything never than 0.6-214@220331 (Linux)

Changelog details: Improved Nvidia MEM Temp reporting (updated nvtool to v1.6.0: added memory temperature for GDDR6X; fixed vendor detection for HBM2 memory)

2. Run this command

That should work. If not perform the upgrade anyway and then run this command.

curl https://raw.githubusercontent.com/cryptopoo/hiveos_gddr6x_temps/main/quick-install.sh | bash

Need more info? Look at the image below. Click at the “>_” part of the screen then type the above command at the “>_” prompt. and let it run for a minute

Refresh the screen.

If you have had to run the command, the display changes might not be permanent and you might have to run it again…or not! The command you executed remains in “Worker Commands” so you can easily run it again.

3. This is what you see

This is my test rig and you can see the RTX 3090 is very naughty. The memory is operating at 106 degrees and it should definitely be less than 100 degrees, some of my 3090s go to 110. I find that all my 3090 models have overheating memory even though it is the same memory that is in my 3080 models.

3. Poor RTX Performance

It’s hard enough to get the right settings, let alone understand if you are one of the many people who have bought new RTX cards that have rubbish thermal pads in them.

This new bit of information should help you understand if your memory is getting hot. Remember that ETH mining is MEM intensive so if you lower the MEM (in my case shown above as 1800) then the temperature should lower as well.

Or you could clean your card, increase the external/internal fans, buy an aircon unit or get someone to maintain your graphics card by changing the pads. this might invalidate the warranty as you a) shouldn’t modify the card, although you can maintain it b) shouldn’t use it for mining!

4. RTX Thermal Throttling

If you think your card is performing slowly and if you also think it’s getting too hot then try running this command in the same way you ran the command above. It takes a minute or so to run.

nvidia-smi -q | egrep “Thermal Slowdown”

Hopefully, the results will look the same as in this image for 3 cards. It shows there is NO hardware or software throttling going on. That’s what you want. If there is throttling then try changing your settings to a commonly used one or get your pads changed.

These pads are the best: Gelid from Amazon.

These are not quite as thermally efficient as Gelid but are cheaper: Snowbabe from Amazon

You absolutely MUST get the right thickness pads. these pads MUST be compressed when you come to close the GPU after you’ve changed them. If there is no compression then there is no heat transfer.