Is this rolling out in waves?
Update doesn’t appear to be catching this.
I have the MSI EdgeExpert variant so hopefully I don’t need to wait for MSI to push something out (can’t imagine I would).
The FE DXG Sparks are pulling the 7.4.0 updates already:
elsaco@spark1:~$ cat /etc/dgx-release
DGX_NAME="DGX Spark"
DGX_PRETTY_NAME="NVIDIA DGX Spark"
DGX_SWBUILD_DATE="2025-09-10-13-50-03"
DGX_SWBUILD_VERSION="7.2.3"
DGX_COMMIT_ID="833b4a7"
DGX_PLATFORM="DGX Server for KVM"
DGX_SERIAL_NUMBER="1923925509726"
DGX_OTA_VERSION="7.3.1"
DGX_OTA_DATE="Wed Nov 19 03:57:20 PM PST 2025"
DGX_OTA_VERSION="7.4.0"
DGX_OTA_DATE="Tue Feb 3 09:06:56 AM PST 2026"
What kernel is it using? Have they finally switched to 6.17?
Yep.
1 Like
The latest kernel is 6.17.0-1004-nvidia
1 Like
Huh, weird. I show as being on 7.4.0, but still on 6.14 kernel. Both on a FE and an EdgeXpert (that’s flashed with the Nvidia image).
Welcome to NVIDIA DGX Spark Version 7.4.0 (GNU/Linux 6.14.0-1015-nvidia aarch64)
Also on OTA 7.4 with the 6.14 kernel still. Looks like the OTA version bump and kernel upgrade are decoupled. The 6.17 kernel packages (linux-image-6.17.0-1008-nvidia) haven’t hit the repos yet — apt-cache depends linux-image-nvidia-hwe-24.04 still points to 6.14.
Guessing NVIDIA will push the kernel as a separate rollout when it’s ready?
Would be nice to have a dev confirm this.
1 Like
@Balaxxe the new kernel is in the repos:
elsaco@spark1:~$ apt search linux-image-6.17.0-1004-nvidia
Sorting... Done
Full Text Search... Done
linux-image-6.17.0-1004-nvidia/noble-updates,noble-security,now 6.17.0-1004.4 arm64 [installed]
Signed kernel image nvidia
linux-image-6.17.0-1004-nvidia-64k/noble-updates,noble-security 6.17.0-1004.4 arm64
Signed kernel image nvidia-64k
There’s even a 64k kernel available!
elsaco@spark1:~$ uname -a
Linux spark1 6.17.0-1004-nvidia #4-Ubuntu SMP PREEMPT_DYNAMIC Tue Nov 18 01:24:05 UTC 2025 aarch64 aarch64 aarch64 GNU/Linux
Also, I upgraded the GPU driver:
elsaco@spark1:~$ nvidia-smi
Tue Feb 3 16:41:33 2026
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 590.48.01 Driver Version: 590.48.01 CUDA Version: 13.1 |
+-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 NVIDIA GB10 On | 0000000F:01:00.0 On | N/A |
| N/A 36C P8 5W / N/A | Not Supported | 1% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
+-----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
| 0 N/A N/A 4100 G /usr/lib/xorg/Xorg 656MiB |
| 0 N/A N/A 4246 G /usr/bin/gnome-shell 333MiB |
| 0 N/A N/A 4498 G /usr/bin/gnome-software 65MiB |
| 0 N/A N/A 6806 G /usr/bin/gnome-text-editor 46MiB |
| 0 N/A N/A 7622 G ...rack-uuid=3190708988185955192 351MiB |
+-----------------------------------------------------------------------------------------+
However, the 590.48.01 driver pulls in dkms and a MOK key must be enrolled in order to have to modules loaded.
elsaco@spark1:~$ dkms status
nvidia/590.48.01, 6.11.0-1016-nvidia, arm64: installed
nvidia/590.48.01, 6.14.0-1015-nvidia, arm64: installed
nvidia/590.48.01, 6.17.0-1004-nvidia, aarch64: installed
I noticed the metapackage (linux-image-nvidia-hwe-24.04) still depends on 6.14.0-1015-nvidia, so the 6.17 kernel won’t pull in via normal apt upgrade yet. I’m waiting for the metapackage to update so it comes through the proper DGX OS upgrade path.
I’d advise against installing the kernel or driver manually – the DGX OS 7.4.0 release notes (Feb 2) specify 580.126.09 for DGX Spark with CUDA 13.0 Update 2, not 590/CUDA 13.1. The release notes also list 6.17.0-1008-nvidia as the target kernel for Spark, not 1004. 1008 is the latest build and is still sitting in the proposed repo. You might be running older/mismatched packages that haven’t been validated together for our hardware. I’d rather wait for everything to land as a coordinated update through the metapackage so the kernel, driver, and CUDA versions all match what NVIDIA tested.
Edit: FYI - Everything updated for me smoothly EXCEPT for the kernel (which seems to be pending official release).
What’s the output of nvcc --version?
Welcome to NVIDIA DGX Spark Version 7.4.0 (GNU/Linux 6.14.0-1015-nvidia aarch64)
[...]
> nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2025 NVIDIA Corporation
Built on Wed_Aug_20_01:57:39_PM_PDT_2025
Cuda compilation tools, release 13.0, V13.0.88
Build cuda_13.0.r13.0/compiler.36424714_0
Can any dev confirm when 6.17 (1008) will be released via OTA?
For those wondering why the 6.17.0-1008-nvidia kernel mentioned in the 7.4.0 release notes hasn’t landed yet, I did some digging.
The kernel is tracked in Ubuntu’s SRU workflow here: Bug #2138765 “noble/linux-nvidia-6.17: 6.17.0-1008.8 -proposed t...” : Bugs : Kernel SRU Workflow
Boot testing, secure boot, signing, and ABI testing all passed. But automated testing and regression testing are marked as failed. Verification testing is still in progress it looks like.
On Feb 3, someone manually added a kernel-block-proposed tag, explicitly holding the release. This was the day after the release notes came out, so they probably found something during final validation (I’m assuming).
To check when it’s ready:
apt-cache depends linux-image-nvidia-hwe-24.04 | grep linux-image
When that shows 6.17.0-1008-nvidia instead of 6.14.0-1015-nvidia, should be safe to upgrade through the normal apt workflow.
Really wished Nvidia would keep us posted on things like this and/or just not release until ready.
1 Like
For anyone who wants to manually upgrade to the official Nvidia 6.17 kernel while waiting for the official release, see below.
The kernel and nvidia modules are in Ubuntu’s proposed repo. You can install them, remove proposed, and future updates will come through normally via apt. This includes any point releases to 1008 (like 1008.9 if they push fixes) as well as 1009 or the official release.
sudo add-apt-repository "deb http://ports.ubuntu.com/ubuntu-ports noble-proposed main restricted universe multiverse"
sudo apt update
sudo apt install linux-image-6.17.0-1008-nvidia linux-modules-6.17.0-1008-nvidia linux-modules-nvidia-580-open-6.17.0-1008-nvidia linux-modules-nvidia-fs-6.17.0-1008-nvidia
sudo add-apt-repository --remove "deb http://ports.ubuntu.com/ubuntu-ports noble-proposed main restricted universe multiverse"
sudo reboot
After reboot, verify with:
uname -r
nvidia-smi
You should see kernel 6.17.0-1008-nvidia and driver 580.126.09.
2 Likes
Hmm. Funny. Here is the situation with my two sparks after a morning update.
uname -r
6.17.0-1008-nvidia
nvidia-smi
Thu Feb 12 19:26:27 2026
±----------------------------------------------------------------------------------------+
| NVIDIA-SMI 590.48.01 Driver Version: 590.48.01 CUDA Version: 13.1 |
±----------------------------------------±-----------------------±---------------------+
Things seems to be working ok without any hiccups so far (for me anyway)
Kernel got upgraded, but I’m still at 580.126.09 drivers
Same here, 6.17.0 kernel and 580.126.09 drivers.
The official release notes confirm that this is the expected combo. It was officially confirmed that 590 was not ready for the GB10 yet.
Oh well. I will stay on 590 as I see no issues yet. So far tested with running few models with vllm, llama.cpp and LocalAI.
I think I know what happened with my sparks and its due to experiencing problem with random reboots for no particular reason on one, I do remember now upgrading to the latest available kernel and drivers in my efforts looking for a fix, which BTW, helped with stability on my spark rebooting constantly after 10-15 minutes uptime.
Sorry about the confusion folks.
Let me know if you would like me to execute any specific tests with this OS/Kernel/Nvidia drivers and I’d more than happy to do so.