I have a DGX Spark at work (a Gigabyte AI TOP ATOM). I use it mainly to run local LLMs for coding, translation and generic technical questions, but I also use it fairly regularly to model semi-complex stuff in Blender, and render long 360° videos.

I really like this little machine: it’s dead quiet, it’s a decently useful AI assistant for “non-vibe coding” - i.e. using AI to code sample code, boilerplates and fix tiny code snippets without letting it do anything complex or have access to any important files - it’s a great Blender box and it’s an okay video renderer for an arm64 machine.

It is fairly disappointing for video rendering, but that’s mainly because I use kdenlive and kdenlive is not great with hardware acceleration, and my 360° videos are too large / non-standard aspect ratio to be rendered in the GPU anyway. Still, it manages to render 4096x2048 videos at around 15 frames per second in CPU-only. Not too shabby.

The box wasn’t cheap though - just north of 5k€. But I suspect a lot of that price is the cost of RAM. Also, I don’t care since it was company money.

But at some point, I’d like to buy one for myself at home: the ability to use it to OCR sensitive paper mail and translate the letters locally without hitting fucking Google or Microsoft servers and sending them private information alone is well worth the money, and I need a new Blender machine at home anyway. But I haven’t done it yet because, well… 5k€ is kind of yikes when I’m the one paying 🙂