Why does NVIDIA keep neglecting Spark with zero-day support

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

Boy – Qwen 3.8-Next looks like an ideal model for two DGX Sparks. Sparse architecture but takes advantage of the generous available memory. They explicitly mention the model in their breathless press release:

"Beyond rack-scale deployment, Qwen3.8-Flash-Next also runs on local NVIDIA hardware, including NVIDIA DGX Station, NVIDIA DGX Spark clusters, "

So why not, you know, publish a configuration that actually works? Particularly in the NVFP4 format that’s perfect for this model.

Now I know I can download some zero star vibe-coded hack of a vllm container with random patches to sort of get it running. Or worse, go down an entirely new rabbit hole with SGLang. But WHY does a company that literally makes so much money they don’t know what to do with it completely abuse its enthusiast base?

2

Just one/two/four/whatever DGX Sparks don’t even occupy a single cell on their revenue dashboard

3

Might be. But Apple seems to think there is a real growth opportunity in local AI. They haven’t been wrong often.

My point is - why bother releasing a hardware platform to neglect it?

4

The answer seems very simple. Unlike Apple, NVIDIA has higher-margin products that need support right now (GB300). Gamers, professionals, and enthusiasts get the hardware, and then they have to make it work with the help of the community (and opus).

5

Can we not use the term, “zero-day” for nvidia not having a working runtime ready for a model’s launch day…

6

Yes, I initially thought it would be talking about a critical vulnerability too.