Agentic DevOps with DGX Spark ?!

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

Hey Fam - What do you think about the prospect of leveraging the DGX Spark for locally hosted Agentic DevOps pipelines?

Specifically, I would like your inputs/discussions on:

  • Open-weight LLMs with tool calling and how local nVidia DGX stack will support it
  • Promises and potential limits for agentic coding and other 24/7 pipelines
  • Your experience with agentic coding so far

Thanks a lot and look forward to reading your inputs.

2

**So far my research has led me to believe that leveraging the DGX Spark for locally hosted Agentic DevOps pipelines is definitely possible, but Iā€™m still looking for solid confirmation.**What are the current bottlenecks ā€” memory, context length, hallucination, or something else?

3

The DGX Spark should definitely support some level of Agentic AI. These days, gpt-oss, qwen3 and mistral-small LLMs are doing OK for tool calls. They may not be as smart as frontier models, but you can get some good results by breaking your problem in smaller steps and finetuning your prompts with specific tools/instructions at each step. We should be able to run those models on DGX Spark (aarch64) with Nvidia NIMs or Ollama/Llama.cpp.

For Coding/DevOps purpose, we need to keep in mind this is an ARM/aarch64 box so check your favourite Docker images, tools and native dependencies to see if they are compatible with this architecture.

4

I believe stack of two DGX Spark will easily handle team of few 13B/34B models task specific , probably the bottleneck will be the time to feed them , and still plenty room to avoid Latency, no trottling i believe and use by multiple people at same time. Please correct me if im wrong .

1 Like