DGX Spark / GB10 FAQ

Accelerated Computing DGX Spark / GB10 User Forum DGX Spark / GB10
1

Initial Setup

Q: I am trying to set up my DGX Spark in appliance mode. I am able to connect to the DGX Spark SSID, but it can’t open the webpage it tries to redirect me to (http://spark-xxxx.local).

To connect to http://spark-xxxx.local your network needs to be configured to enable communication between devices on the local network/subnet.

Windows:

  1. Set WiFi network as private:

    1. Go to “Settings” > “Network & Internet” > “WiFi”

    2. Open the properties of your WiFi network

    3. Set the network profile type to “Private Network”

  2. Enable network discovery and file and printer sharing:

    1. Go to “Settings” > “Network & Internet” > “Advanced Network Settings” > “Advanced sharing settings”

    2. Open the settings for “Private networks”

    3. Turn On “Network Discovery” and “File and printer sharing”

Mac:

By default, macOS does not allow third party applications to communicate with devices on your local network / subnet. Please ensure that local network is enabled for your browser: Control access to your local network on MacOS

Q: I don’t see the Spark SSID.

The DGX Spark will only broadcast the SSID before the initial setup is complete. If you have already completed the setup, this is expected behavior. If you have not completed the setup:

  1. Confirm the DGX Spark is powered on:

    1. There are no indicator lights on the DGX Spark Founders Edition

    2. To detect if the device is ON you can ping the Spark using the hostname provided in the Quick Start Guide ping spark-xxxx

      1. If you receive a response but still do not see the SSID:

        1. Turn off the WiFi adapter on your client machine. Wait 5 seconds and then turn it back on.

        2. Try power cycling the DGX Spark. Hold the power button on the back of the device for five seconds. Wait five seconds, then press and release the power button to turn the DGX Spark back on

        3. If the issue persists please share your experience on the developer forum. You can also use desktop mode (setup with keyboard, mouse and monitor) to set up the device.

      2. If there is no response from the DGX Spark, hold the power button on the back of the device for five seconds. Wait five seconds, then press and release the power button to turn the DGX Spark back on.

Q: Does my DGX Spark power on as soon as I plug it in?

Yes. Your DGX Spark will turn on as soon as you plug it in. To power-off, hold the power button on the back of the device for five seconds. For full details see the Quick Start Guide.

Using your DGX Spark

Q: I am unable to SSH to my DGX Spark from my third party terminal application on my macOS device.

By default, macOS does not allow third party applications to communicate with devices on your local network/subnet. Please ensure that local network is enabled for your third party terminal: Control access to your local network on MacOS.

Q: When I run nvidia-smi to see the memory usage it says “Not Supported”.

This is expected behavior. The DGX Spark has a unified memory architecture. NVIDIA-SMI only reports memory utilization when there is a dedicated GPU VRAM. To check the memory usage you can use the following commands top, htop or free, or using the DGX Dashboard https://localhost:11000

Q: My application is experiencing memory issues even though I am within the memory capacity of the DGX Spark.

The DGX Spark uses a unified memory architecture, a powerful architecture that allows for dynamic memory sharing between the GPU and CPU. Many applications are still updating to take advantage of this new architecture.

If you experience memory issues, despite being within the memory capacity of the DGX Spark, you can check if some memory is being held in the cache by running the free -h and checking the buff/cache field. If you see memory is being stored in cache, you can resolve this issue by flushing the buffer cache manually:

sudo sh -c 'sync; echo 3 > /proc/sys/vm/drop_caches'

Q: My display monitor is unable to detect my DGX Spark in desktop mode

For the best display experience we recommend using the USB-C to DP or USB-C to USB-C cables that are shipped with the monitor or VESA certified cables.

If the monitor does not light up when connected using USB-C to DP or USB-C to USB-C cables please troubleshoot using the following steps:

For setups where your monitor is connected directly to your DGX Spark:

  1. Use the HDMI port of the DGX Spark using an HDMI to HDMI cable.

  2. Unplug the display monitor and plug it back in.

  3. Check if your monitor has an MST feature. If so, disable it using the monitor OSD menu.

  4. Check if your monitor has a power-saving feature in the monitor OSD menu. Ensure that it is set to “normal level” or “deep level.”

  5. Lower the resolution and the refresh rate to 4K/120 Hz or lower.

For setups where you are using a dock or hub:

  1. Connect the monitor directly to the DGX Spark using either HDMI, USB-C to DP or USB-C to USB-C.

Please note tunneling or monitors connected in a daisy chain with a thunderbolt monitor is not supported.

Q: I am experiencing issues with NVIDIA Container Runtime for Docker

Instructions for installing and using Docker with the NVIDIA Container Toolkit are available in the DGX Spark User Guide.

Q: Are NIMs available on DGX Spark?

Available NIMs, and Playbooks for running them, can be found on build.nvidia.com.

Q: I deleted my device from NVIDIA Sync. When I try to re-add the device I am getting an error message “Host already exists”.

NVIDIA Sync maintains an ssh config file that is imported into your default ssh config file. The alias may still be present in the config file. To resolve this issue, check the following paths for the SSH config file and remove the hostname you are trying to re-add if it exists:

  • Windows: C:\Users\<username>\AppData\Local\NVIDIA Corporation\Sync\config\ssh_config

  • MacOS: /Users/<username>/Library/Application Support/NVIDIA/Sync/config/ssh_config

  • Linux: /home/<username>/.config/NVIDIA/Sync/config/ssh_config

Q: Is GPUDirect RDMA supported on DGX Spark?

DGX Spark SoC is characterized by a unified memory architecture.

For performance reasons, specifically for CUDA contexts associated to the iGPU, the system memory returned by the pinned device memory allocators (e.g. cudaMalloc) cannot be coherently accessed by the CPU complex nor by I/O peripherals like PCI Express devices.

Hence the GPUDirect RDMA technology is not supported, and the mechanisms for direct I/O based on that technology, for example nvidia-peermem (for DOCA-Host), dma-buf or GDRCopy, do not work.

A compliant application should programmatically introspect the relevant platform capabilities, e.g. by querying CU_DEVICE_ATTRIBUTE_GPU_DIRECT_RDMA_SUPPORTED (related to nv-p2p kernel APIs) or CU_DEVICE_ATTRIBUTE_DMA_BUF_SUPPORT (related to dma-buf), and leverage an appropriate fallback.

For example, for Linux RDMA applications based on the ib verbs library, we suggest to allocate the communication buffers with the cudaHostAlloc API and to register them with the ib_reg_mr function.

Q: How do I setup my DGX Spark when I move locations?

  1. Connect Spark to router via Ethernet

  2. Try to connect with spark-hostname.local, if it works go to step 4

  3. View devices connected to router and find Spark (spark-hostname) IP
    a. Method may vary, you can use your router administrative panel or you may be able to connect directly to your spark by using the hostname if DNS works

  4. Connect to Spark via SSH or NV Sync app to open command line

  5. Use nmcli command to configure WiFi
    sudo nmcli d wifi connect <wifi_name> password <password>

  6. If connecting by hostname does not work, find IP address for WiFi interface
    ip -f inet a show wlP9s9

  7. Disconnect Spark from ethernet and use hostname or WiFi IP to connect

2 Likes