[Blog] NVIDIA DGX Spark (#3298)

Author: peterschmidt85

docs/blog/posts/nvidia-dgx-spark.md

@@ -0,0 +1,132 @@
+---
+title: "Orchestrating workloads on NVIDIA DGX Spark"
+date: 2025-11-14
+description: "TBA"
+slug: nvidia-dgx-spark
+image: https://dstack.ai/static-assets/static-assets/images/nvidia-dgx-spark.png
+# categories:
+#   - Benchmarks
+---
+
+# Orchestrating workloads on NVIDIA DGX Spark
+
+With support from [Graphsignal :material-arrow-top-right-thin:{ .external }](https://x.com/GraphsignalAI/status/1986565583593197885){:target="_blank" }, our team gained access to the new [NVIDIA DGX Spark :material-arrow-top-right-thin:{ .external }](https://www.nvidia.com/en-us/products/workstations/dgx-spark/){:target="_blank"} and used it to validate how `dstack` operates on this hardware. This post walks through how to set it up with `dstack` and use it alongside existing on-prem clusters or GPU cloud environments to run workloads.
+
+<img src="https://dstack.ai/static-assets/static-assets/images/nvidia-dgx-spark.png" width="630"/>
+
+<!-- more -->
+
+If DGX Spark is new to you, here is a quick breakdown of the key specs.
+
+* Built on the NVIDIA GB10 Grace Blackwell Superchip with Arm CPUs.
+* Capable of up to 1 petaflop of AI compute at FP4 precision, roughly comparable to RTX 5070 performance.
+* Features 128GB of unified CPU and GPU memory enabled by the Grace Blackwell architecture.
+* Ships with NVIDIA DGX OS (a tuned Ubuntu build) and NVIDIA Container Toolkit.
+
+These characteristics make DGX Spark a fitting extension for local development and smaller-scale model training or inference, including workloads up to the GPT-OSS 120B range.
+
+## Creating an SSH fleet
+
+Because DGX Spark supports SSH and containers, integrating it with dstack is straightforward. Start by configuring an [SSH fleet](../../docs/concepts/fleets.md#ssh-fleets). The file needs the hosts and access credentials.
+
+<div editor-title="fleet.dstack.yml">
+
+```yaml
+type: fleet
+name: spark
+
+ssh_config:
+  user: devops
+  identity_file: ~/.ssh/id_rsa
+  hosts:
+    - spark-e3a4
+```
+
+</div>
+
+The `user` must have `sudo` privileges.
+
+Apply the configuration:
+
+<div class="termy">
+
+```shell
+$ dstack apply -f fleet.dstack.yml
+
+Provisioning...
+---> 100%
+
+ FLEET  INSTANCE  GPU     PRICE  STATUS  CREATED 
+ spark  0         GB10:1  $0     idle    3 mins ago      
+```
+
+</div>
+
+Once active, the system detects hardware and marks the instance as `idle`. From here, you can run
+[dev environments](../../docs/concepts/dev-environments.md), [tasks](../../docs/concepts/tasks.md), 
+and [services](../../docs/concepts/services.md) on the DGX Spark fleet, the same way you would with other on-prem or cloud GPU backends.
+
+## Running a dev environment
+
+Example configuration:
+
+<div editor-title=".dstack.yml">
+
+```yaml
+type: dev-environment
+name: cursor
+
+image: lmsysorg/sglang:spark
+
+ide: cursor
+
+resources:
+  gpu: GB10
+
+volumes:
+  - /root/.cache/huggingface:/root/.cache/huggingface
+
+fleets: [spark]
+```
+
+</div>
+
+We use an [instance volume](../../docs/concepts/volumes.md#instance-volumes) to keep model downloads cached across runs. The `lmsysorg/sglang:spark` image is tuned for inference on DGX Spark. Any Arm-compatible image with proper driver support will work if customization is needed.
+
+Run the environment:
+
+<div class="termy">
+
+```shell
+$ dstack apply -f .dstack.yml
+
+ BACKEND       GPU     INSTANCE TYPE  PRICE  
+ ssh (remtoe)  GB10:1  instance       $0     idle
+
+Submit the run cursor? [y/n]: y
+ 
+ #  NAME    BACKEND       GPU     PRICE  STATUS   SUMBITTED
+ 1  cursor  ssh (remote)  GB10:1  $0     running  12:24
+
+Launching `cursor`...
+---> 100%
+
+To open in VS Code Desktop, use this link:
+  vscode://vscode-remote/ssh-remote+cursor/workflow
+```
+
+</div>
+
+Workloads behave exactly like they do on other supported compute targets. You can use DGX Spark for fine tuning, interactive development, or model serving without changing workflows.
+
+!!! info "Aknowledgement"
+    Thanks to the [Graphsignal :material-arrow-top-right-thin:{ .external }](https://graphsignal.com/){:target="_blank"} team for access to DGX Spark and for supporting testing and validation. Graphsignal provides inference observability tooling used to profile CUDA workloads during both training and inference.
+
+## What's next?
+
+1. Read the [NVIDIA DGX Spark in-depth review :material-arrow-top-right-thin:{ .external }](https://lmsys.org/blog/2025-10-13-nvidia-dgx-spark/){:target="_blank"} by the SGLang team.
+2. Check [dev environments](../../docs/concepts/dev-environments.md), 
+    [tasks](../../docs/concepts/tasks.md), [services](../../docs/concepts/services.md), 
+    and [fleets](../../docs/concepts/fleets.md)
+3. Follow [Quickstart](../../docs/quickstart.md)
+4. Join [Discord :material-arrow-top-right-thin:{ .external }](https://discord.gg/u8SmfwPpMd){:target="_blank"}