NVidia Workbench install failed on download of WSL distro

Please describe your issue or request:

I’m trying to install AI Workbench on my Windows 11 desktop (25H2).

I have virtualization installed

I have the latest WSL2 installed (I’m already running an Ubuntu and docker-desktop images)

I have Docker Desktop installed

I have UAC enabled

When trying to install, it gets to the point where it’s trying to download the wsl distro and retries and fails saying there’s a connection error.

I tried turning firewall off - no change.

I tried running as admin and as normal user (with UAC confirmation)

I also tried manually downloading and installing the AI-Workbench distro. The installer saw it, but couldn’t configure it, probably because I manually installed it and didn’t put any specific setup parameters into it.

I also just tried manually downloading ubuntu-jammy-wsl-amd64-wsl.rootfs.tar.gz and putting it in

I’ve attached the install logs. I’m at a loss for how to fix this.

Also, I’ve attached the logs from after I manually installed the AI-Workbench distro (ubuntu-jammy-wsl-amd64-wsl.rootfs.tar.gz).

Please tick the appropriate box to help us categorize your post
🗹 Bug or Error
Feature Request
Documentation Issue
Other

after_manual_wsl_distro_install_logs.txt (706.9 KB)

main.log (61.8 KB)

Bumping this.. is there anyone from the NVidia team who can assist? I also emailed your support and have had no response.

Your installer is indeed failing to download the Ubuntu rootfs:

[info]  (check-if-exists)   C:\Users\tim\AppData\Local\Temp\ubuntu-jammy-wsl-amd64-wsl.rootfs.tar.gz
[error] (check-if-exists)   checkIfExists ENOENT: no such file or directory, access 'C:\Users\tim\AppData\Local\Temp\ubuntu-jammy-wsl-amd64-wsl.rootfs.tar.gz'
[info]  (download-util)     downloadFile() https://cloud-images.ubuntu.com/wsl/releases/jammy/current/ubuntu-jammy-wsl-amd64-wsl.rootfs.tar.gz C:\Users\tim\AppData\Local\Temp\ubuntu-jammy-wsl-amd64-wsl.rootfs.tar.gz distro_download_progress
[info]  (download-util)     ubuntu-jammy-wsl-amd64-wsl.rootfs.tar.gz: STARTED
[info]  (download-util)     ubuntu-jammy-wsl-amd64-wsl.rootfs.tar.gz: RETRY
[info]  (download-util)
                            RetryAttempt: 1/5,
                            StartsOn: 5 secs,
                            Reason:

What you could do is download https://cloud-images.ubuntu.com/wsl/releases/jammy/current/ubuntu-jammy-wsl-amd64-wsl.rootfs.tar.gz and drop it in C:\Users\tim\AppData\Local\Temp\ so the installer will find it there and not try to download it.

I seem to have the same issue.

Tried:

  • Closing AI Workbench
  • downloading the .gz file and put it in the Temp folder
  • Restarted AI Workbench
  • the .gz file is removed from the Temp folder
  • same symptoms and error in AI Workbench.

This is occurring for me on two different comptuers - one PC with Windows 11 25H2 , one Lenovo P1 Gen 6 laptop with Windows 11 24H2. The same issue occured prior to reimaging one of the PCs too.

I had English Australian set as primary language. Tried chaging to US English - this did not help.

Copying the wsl file to temp doesn’t work, as the installer seems to delete it as soon as it starts.

This is occurring for me on two different comptuers - one PC with Windows 11 25H2 , one Lenovo P1 Gen 6 laptop with Windows 11 24H2. The same issue occured prior to reimaging one of the PCs too.

I had English Australian set as primary language. Tried chaging to US English - this did not help.

I’m using English Australia too. I also tried English US and it didn’t work. I’ve hit a dead-end here. Mine is also a PC with W11 25H2.

I’m having the same issue. Using windows 11 Business 25H2.

I’ve encountered this issue for many months.

Need to get it fixed now as I have purchased a DGX Spark.

Hoping Nvidia can help with a fix.

I need it for the Spark too. Right now it’s a paperweight that looks like a DGX Spark. I need to run headless with AI Workbench tools to get any real value out of it for our R&D.

Hello,

The install process has been overhauled in the latest release of AI Workbench, could you kindly try it and let us know if you experience the same issue?

Download: NVIDIA AI Workbench

I’m now having an issue with the latest version. I’m using the Add Remote Machine option (from Windows 11) and I get the following error;

{“level”:“error”,“pid”:73540,“error”:“cannot connect to ssh-agent: problem connecting to OpenSSH Authentication Agent, ensure the service is running”,“isWrapped”:false,“isInteractive”:false,“engine”:“json”,“detail”:“detail”,“time”:“2025-11-28T02:56:41+11:00”,“message”:“Failed to get context SSH fingerprints.”}

I have my pub key in authorized_keys and also the remote host fingerprint in my known_hosts and it still fails. The command it’s trying to run is:

nvwb-cli.exe -o json create context --accept-ssh-fingerprints home <<HIDDEN>> --description “Spark1” --context-workbench-dir /home/tim/.nvwb --ssh-port 22 --ssh-key-path “C:\Users\tim\.ssh\id_ed25519.pub” --ssh-username <<HIDDEN>>

To add to this.. I can ssh into it with no problem (using my pub key).. so I’m not sure where it’s trying to get the context fingerprint??

Hello, good to hear you’re installed. Can you please first try cycling your WSL distro with:

# shutdown distro
wsl -t NVIDIA-Workbench

# start distro
wsl -d NVIDIA-Workbench

# try connecting again

If this still doesn’t work, please create a support bundle and attach it here for us to inspect additional relevant logs.

I am still unable to install the AI Workbench WSL image locally.

Download still failing with new installer.

The image can be downloaded and imported to WSL manually. However that does not help, as folders and “workbench” user are not yet present in the image.

I’ve emailed the log to the support email address. If someone could please take a look and email me back that would be much appreciated.

I’ve since manually installed Docker 1st.. that did not help.

I’m not sure I see your emailed log. Can you please confirm the send date and subject? We request a support bundle zip because it contains various logs, sometimes a single log isn’t enough for us to determine an issue.

Never mind, I have found it.