I noticed that after updating 2 updates ago from the system from the NVIDIA dashboard that the Spark was incredibly hot when not logged-in, but cooled down as soon as I logged-in. I had Claude Coder see what changed, and it applied a fix, which seemed to work until the next update. Claude tried another fix, but that hasn’t worked. Called Nvidia customer support and advised I needed to post here for customer assistance. To that end I’m attaching the Claude diagnostic log…please advise what update I need to install to fix this issue…
Spark_GPU_Overheating_LogMDB.txt (1.8 KB)
I spoke to engineering about your issue and they asked for some more information
- Repro steps or setup steps so they can try a local reproduction
- When issue happens, start a SSH connection and collect nvidia-smi to check the GPU status
- When issue happened, start a SSH connection and execute sensors command to collect temperature data (need to apt install lm-sensors first)
- For isolation, plug out ethernet, wait for a while and check if issue happen in order to avoid unexpected connection
The problem is the issue happens before I logon to the Spark, and goes away once I’m logged-in. Claude had me logon from another Machine, but by logging on, the issue went away. He found some logs on the Spark, which were reflected in the txt file I uploaded to the forum.
The curious thing is my colleague logged–in from Berlin to do some work, then logged–out—and after he logged–out, the Spark stayed in ready mode without overheating for a day.
The other curiosity is that I didn’t have this issue until after updating the Spark 3 updates ago. Before that, the unit stayed in standby/pre-login mode with no overheating issues.
Please advise how to obtain the data logs from the Spark, and how to invoke the Nvidia-SMI without logging into the Spark.
Many thanks—Hopkins Stanley
Hopkins Stanley
exnihiloheide@gmail.com
Ex nihilo, exc.
We you said “logged in” do you mean logging into the Gnome UI or ssh?
I mean turning the Spark on and arriving at the logon screen, but not logging-in. It’s when the Spark is turned on, but not logged into and is idle, waiting for someone to remotely log in. The Spark happily sat there in this pre-login state, waiting for my colleague (or myself) to remotely log into from Berlin.
But after installing Nvidia’s Spark software update somewhere around 2/6/2026, the unit starting getting as hot as when it’s cranking on a ComfyUI video generation process, actually warmer. As, per the txt file I uploaded to the forum, Claude found the system log files showed the GPU was now running at full-tilt boogie during the Gnome pre-logon state, and noticed something had changed in the config.sys file after the 2/2/2026 NVidia system update.
Unfortunately, I’m not a programmer, but a phenomenologist…and, as such, I’m unclear on how you SSH into the device to initiate the NVidia-SMI command with Gnome UI when ssh into the device, or for that matter do anything without having access to the OS’s commands to initiate reading/writing/copying stuff?
The weird thing is the unit turns into a toaster on powering up, but cools off once you enter your password and can use the machine. It makes me think something happened in the UEFI procedure perhaps?
Best and many thanks…Hopkins Stanley
Hopkins Stanley
exnihiloheide@gmail.com
Ex nihilo, exc.
You will have to login to run nvidia-smi. I would say collect the sensor data at least, hopefully it will show the unit at a higher temperature right when you login. Also, please disconnect ethernet/wifi for isolation.
So…I logged in and obtained the Nvidia-smi readings immediately after logging in, both with and without Internet/Wifi, then + 30 minutes post-logging in, and learned that:
1: Logging in with Internet/Wifi shows immediate GPU temp as 53 °C, and it is quite hot to the touch. [3.5.2026Nvidia-SmiWOInternetWifi.txt]
2. Logging in without Internet/Wifi shows immediate GPU temp as 56 °C, and is quite »toasty« to the touch. [3.5.2026Nvidia-SmiWOInternetWifi.txt]
3. Waiting 30 minutes post-logging in, shows GPU temp is down to 42 °C, and much cooler. [3.5.2026Nvidia-SmiWOInternetWifiPost30min.txt]
(Attachment attachment.tiff is missing)
(attachments)
3.5.2026Nvidia-SmiWInternetWifi.txt (1.94 KB)
3.5.2026Nvidia-SmiWOInternetWifi.txt (1.91 KB)
3.5.2026Nvidia-SmiWOInternetWifiPost30min.txt (1.94 KB)
Spark_GPU_Overheating_2026-02-20_GDM_WaylandFix.txt (1.87 KB)
Spark_GPU_Overheating_2026-02-26_GDMMDB.txt (1.89 KB)
Hi…I keep trying to send you the log files, but your system won’t accept them as I received these two messages after trying twice:
First this:
(Attachment attachment.tiff is missing)
Hi…I keep trying to send you files, but the email gets rejected saying first your e-mail isn’t available, then after trying again that the system doesn’t except txt files, among other types…
Please advise…
Hopkins Stanley
exnihiloheide@gmail.com
Ex nihilo, exc.
Third try sending this…
So…I logged in and obtained the Nvidia-smi readings immediately after logging in, both with and without Internet/Wifi, then + 30 minutes post-logging in, and learned that:
1: Logging in with Internet/Wifi shows immediate GPU temp as 53 °C, and it is quite hot to the touch. [3.5.2026Nvidia-SmiWOInternetWifi.txt]
2. Logging in without Internet/Wifi shows immediate GPU temp as 56 °C, and is quite »toasty« to the touch. [3.5.2026Nvidia-SmiWOInternetWifi.txt]
3. Waiting 30 minutes post-logging in, shows GPU temp is down to 42 °C, and much cooler. [3.5.2026Nvidia-SmiWOInternetWifiPost30min.txt]
I’m also attaching the Claude Dividuum’s Diagnostic logs 14.2.2025 and 20.2.2025.
Best and thanks…Hopkins
(Attachment attachment.tiff is missing)
Now I’ve tried 4 times trying to respond and send you the requested data, but your system keeps rejecting it. Please advise…Hopkins Stanley
Hopkins Stanley
exnihiloheide@gmail.com
Ex nihilo, exc.
None of those temperature readings are unusually hot. When it is really under load it exceeds 80C.
You can use the forum UI to send me a private message with the files attached.
But, that’s not even logged into…and why does the temperature go back down to normal after I login…like it did until the February update? Given that the unit is on 24/7 in non-logged-in standby mode, it makes the room a bit hot…
Hopkins Stanley
exnihiloheide@gmail.com
Ex nihilo, exc.
GDM might be pinning a CPU or keeping it from entering lower power states. Use ssh to log into the machine and run something like HTOP and see what the CPU activity is.