I cannot get the network interface “up” on the Ethernet port of my DGX Spark computer. I confirmed the CAT6 cable and the switch port (by another DGX Spark computer), tested with point to point connections on both units, but the interface never becomes active on enP7s7.
Can you generate an nvidia-bug-report while the cable is plugged in?
nvidia-bug-report.log.gz (774.5 KB)
Please see attached.
Thank you for the info. Can you also run sudo ethtool enP7s7?
This will tell us if the link is active. If you see Link detected: no, try running sudo ip link set enP7s7 up and checking again with ethtool
Thank you. I executed the commands, please see below.
fabric@spark-7c26:~$ sudo ethtool enP7s7
[sudo] password for fabric:
Settings for enP7s7:
Supported ports: [ TP ]
Supported link modes: 10baseT/Half 10baseT/Full
100baseT/Half 100baseT/Full
1000baseT/Full
10000baseT/Full
2500baseT/Full
5000baseT/Full
Supported pause frame use: Symmetric Receive-only
Supports auto-negotiation: Yes
Supported FEC modes: Not reported
Advertised link modes: 10baseT/Half 10baseT/Full
100baseT/Half 100baseT/Full
1000baseT/Full
10000baseT/Full
2500baseT/Full
5000baseT/Full
Advertised pause frame use: No
Advertised auto-negotiation: Yes
Advertised FEC modes: Not reported
Speed: Unknown!
Duplex: Unknown! (255)
Auto-negotiation: on
Port: Twisted Pair
PHYAD: 0
Transceiver: internal
MDI-X: Unknown
Supports Wake-on: pumbg
Wake-on: g
Current message level: 0x00000033 (51)
drv probe ifdown ifup
Link detected: no
fabric@spark-7c26:~$
fabric@spark-7c26:~$ sudo ip link set enP7s7 up
fabric@spark-7c26:~$
fabric@spark-7c26:~$ ip link show dev enP7s7
2: enP7s7: <NO-CARRIER,BROADCAST,MULTICAST,UP> mtu 1500 qdisc mq state DOWN mode DEFAULT group default qlen 1000
link/ether 4c:bb:47:2f:7c:26 brd ff:ff:ff:ff:ff:ff
altname enP7p1s0
fabric@spark-7c26:~$
fabric@spark-7c26:~$ sudo ethtool enP7s7
Settings for enP7s7:
Supported ports: [ TP ]
Supported link modes: 10baseT/Half 10baseT/Full
100baseT/Half 100baseT/Full
1000baseT/Full
10000baseT/Full
2500baseT/Full
5000baseT/Full
Supported pause frame use: Symmetric Receive-only
Supports auto-negotiation: Yes
Supported FEC modes: Not reported
Advertised link modes: 10baseT/Half 10baseT/Full
100baseT/Half 100baseT/Full
1000baseT/Full
10000baseT/Full
2500baseT/Full
5000baseT/Full
Advertised pause frame use: No
Advertised auto-negotiation: Yes
Advertised FEC modes: Not reported
Speed: Unknown!
Duplex: Unknown! (255)
Auto-negotiation: on
Port: Twisted Pair
PHYAD: 0
Transceiver: internal
MDI-X: Unknown
Supports Wake-on: pumbg
Wake-on: g
Current message level: 0x00000033 (51)
drv probe ifdown ifup
Link detected: no
fabric@spark-7c26:~$
@aniculescu I posted the outputs from ethtool yesteday. I wanted to ask if you have any feedback about them.
I’m not sure if your issue is a bug. If you turn off and turn on your DGX without rebooting, see if the Ethernet port works. There is a known bug where the system might mix up the NIC model after a reboot.
You may check this ( Realtek r8127 Ethernet NIC disappears on every reboot, only recovers with full power cycle ) if you encounter the bug
The bug report and ethtool output indicate that the interface never establishes a physical link.
ethtool shows:
- Speed: Unknown!
- Duplex: Unknown! (255)
- Link detected: no
The interface is up but remains in NO-CARRIER with auto-negotiation enabled, which means no link signal is being detected.
From the driver side:
r8127 0007:01:00.0: invalid VPD tag 0x00 (size 0) at offset 0; assume missing optional EEPROM
This indicates missing optional VPD data but does not directly explain link failure.
Interrupt activity provides an additional signal:
- All 17 MSI-X vectors for enP7s7 show zero counts
The driver initialized and allocated resources, but the device never generated interrupts, consistent with a link that never came up.
MDI-X state is also reported as Unknown, so auto-crossover did not resolve a connection.
With the cable confirmed working on another system, the issue appears localized to this unit.
At this point, the logs show that:
- the driver loads successfully
- the interface is created
- but no physical link is ever established
Further determination (software vs hardware path) would require vendor-level validation.
Thanks for the logs. To confirm, you said you only see this happening on one Spark? Have you tried connecting the Ethernet ports of two Sparks directly together with no switch?
I have just recieved a spark today, it did the out-of-box update over the ethernet port (from the setup process in the manual) and then rebooted and the ethernet port hasnt worked since. Neither the onboard RJ45 or a USB-RJ45 dongle that network manager recognises seem to acknowledge that there is a carrier present anymore.
Hate to be “that guy”, but seems I have this issue. I’ve swapped cables to known working ones, and the OOB experience worked fine but this is a problem.
Yes, I connected the Ethernet ports of the two Sparks directly together with no switch in between, and still there is no link.
ethtool.txt (2.3 KB)
nvidia-bug-report.log.gz (397.9 KB)
Here are the same tests mentioned above from my one. Happy to add an extra data point. I dont have multiple sparks, but have tried connecting it to many different things using many different cables (ex-DC engineer, so there are plenty around the place) and no dice.
Quick update — additional reports show the same pattern.
| Unit | Driver | Kernel | VPD error | MSI-X enP7s7 | Link |
|---|---|---|---|---|---|
| Working unit | 580.142 | 6.17.0-1014 | Present | Active | Up |
| Unit A | 580.142 | 6.17.0-1014 | Present | All zero | Down |
| Unit B | 580.95.05 | 6.14.0-1015 | Present | All zero | Down |
| Unit C | 580.95.05 | 6.14.0-1015 | Present | All zero | Down |
Working and non-working units share the same driver/kernel combinations, and the VPD message appears on all systems.
The consistent signal across failing units remains: no MSI-X interrupt activity after initialization, and no link established.
Hi @mcevik, please submit an RMA request and refer to this comment for RMA approval
Can you please provide information about the platform that I can submit the RMA request? I tried ttps://enterprise-support.nvidia.com however it did not work for me. I also checked the consumer support page for DGX Spark, but I could not find the right place to submit an RMA request.
You can open a ticket with Customer support on the Spark Customer Support website: NVIDIA DGX Spark Support
@aniculescu Unfortunately, I could not find a link that I can open a ticket with Customer support on the link that you indicated. In order to be specific, I’m attaching a screenshot that shows my view of the webpage and I tried the links under “Hardware Support” section. Those links open a page with an error as “HTTP header value exceeds the configured limit of 8192 characters”. Can you please indicate a specific point that I can access your portal to open a case (and request RMA) ? I will also appreciate if you confirm (on your side first) that the page (that you might indicate) works as expected.
Please select the “live chat” option, then ask to speak to an agent from there.
Or you can “open a ticket” and transact over email.