NVIDIA Sync on macOS consistently fails with problem reading key file: open: no such file or directory when connecting to a DGX Spark, even though SSH works perfectly outside of Sync. This appears to be a macOS file access / sandbox issue within NVIDIA Sync.
Environment
macOS (latest version as of Jan 2026)
NVIDIA Sync (latest available build)
DGX Spark / GB10
Same local network (LAN)
What Works
DGX Spark is healthy and reachable
SSH works 100% reliably from Terminal:
ssh <username>@<DGX_IP>
SSH works with:
Password authentication
Key-based authentication
~/.ssh/authorized_keys on the DGX is correct
Permissions verified:
~/.ssh → 700
authorized_keys → 600
Keys are readable by the OS and usable by OpenSSH
This confirms the DGX, SSH service, keys, and network are all functioning correctly.
What Fails
In NVIDIA Sync, pressing Connect always produces:
Unexpected Error
problem reading key file:
open: no such file or directory
This occurs:
After successful device provisioning
After deleting ~/Library/Application Support/NVIDIA/Sync
After reinstalling NVIDIA Sync
After regenerating SSH keys
After manually browsing and selecting the SSH key file
Even when the key is placed outside ~/.ssh (e.g. ~/Downloads/)
The file exists and is readable by the system, but Sync fails to open it.
Steps Already Attempted
Full Sync reset and reinstall
Device removed and re-added multiple times
New SSH key generation
Key copied to non-.ssh directories
Explicit file picker selection inside Sync
macOS reboot
Verified permissions (chmod 600)
Confirmed SSH works immediately outside Sync
Issue persists every time.
Likely Root Cause
This appears to be a macOS sandbox / entitlement issue where NVIDIA Sync does not retain permission to access user-selected key files, even when chosen via the file picker.
Terminal and OpenSSH can read the key, but NVIDIA Sync cannot.
Question
Is this a known NVIDIA Sync issue on macOS, and is there:
A required permission or entitlement that Sync is missing?
A known workaround?
Or confirmation that Sync currently has limited SSH key support on recent macOS versions?
At this point, all client-side troubleshooting has been exhausted.
Hi, can you send me your nvsync logs so I can help better triage this? You should be able to find it at /Users/<username>/Library/Application Support/NVIDIA/Sync/logs/
Hi, I am having the same issue along with the Gb10 Chip not being recognized. It took an immense amount of time to set everything up and I am still having issues with getting the GB10 chip to turn on and use CUDA, and it is overheating. Who do I speak with about the warranty on the DGX Spark?
I am having engineering look at the logs. In the meantime, can you give some more information? Which SSH key are you referring to when you say
After regenerating SSH keys
After manually browsing and selecting the SSH key file
Even when the key is placed outside ~/.ssh (e.g. ~/Downloads/)
NVIDIA Sync creates its own SSH key to communicate with the Spark which it does not keep in the .ssh folder. Also, how are you selecting the SSH key? NVIDIA Sync does not have a file picker to manually select the ssh key. If you have any screenshots or more information that would be very useful.
Yes, I think there’s some terminology mismatch here, so let me clarify exactly what I meant.
When I said “regenerating SSH keys,” I was referring to my standard user SSH keypair that I generate locally and can use successfully from Terminal (for example via ssh -i <key> user@<spark-ip>). I tested with the private key in ~/.ssh/ and also in places like ~/Downloads/ just to rule out any path or permissions assumptions.
When I said “manually browsing/selecting the SSH key file,” I meant the NVIDIA Sync connection setup flow where it prompts for SSH credentials and has an “existing key / key path” style configuration. If there truly is no file picker on your end, then I’m probably describing it poorly, but in my flow there is still a step where I’m either providing a key path or choosing the “use existing key” option.
I understand NVIDIA Sync may generate its own internal key for Spark communication. My issue is that even when I try to use my own existing key, it’s not clear whether Sync is actually using it versus an internally generated key, and the connection behavior is inconsistent.
To help engineering, I can send:
screenshots of the exact NVIDIA Sync screens/fields I’m referencing,
the exact key filename + location used in each test, and
the Spark IP/host string and the full error output from Sync (plus a matching Terminal SSH test).
If you want, I can also do a quick screen recording of the full setup flow end to end.
the NVIDIA Sync connection attempt failing with problem reading key file: open: no such file or directory, and
the UI state during the connection attempt.
This is the behavior I was referring to when mentioning SSH keys. Even though NVIDIA Sync may generate its own internal key, the error suggests it is attempting to read a key file path that either does not exist or is not accessible. There does not appear to be a visible file picker in the UI, so it’s unclear which key or path Sync is actually attempting to use at connect time.
Let me know if you’d like a terminal SSH verification or a short screen recording of the full connection flow.
I just want to be clear, NVIDIA Sync and AI Workbench are different applications. AI Workbench allows modifying the ssh key being used, NVIDIA Sync does not.
Please check if an ssh key was succesfully generated by NVIDIA Sync by checking /Users/<username>/Library/Application Support/NVIDIA/Sync/config/nvsync.key
