# NVIDIA DGX Spark

170 Hacker News commenters recommended NVIDIA DGX Spark and 121 criticized it, November 2025 to November 2026. It was recommended in 12 of 13 months; owners who said how long report a median of 0.5 years (16 owners).

**Verdict:** A quiet, compact 128GB box with fast prompt processing for running large models locally, fine-tuning and keeping data private, though its low memory bandwidth slows token generation.

## Why people recommend it

- Runs large models in 128GB (6 comments): "I've found it very useful for running big models, but it's not a screaming powerhouse in terms of raw compute." ([source](https://news.ycombinator.com/item?id=48426697)) "I can get something like 55 tokens a sec running a larger model like gpt-oss-120b-Q8_0 on the DGX Spark." ([source](https://news.ycombinator.com/item?id=47473631)) "mine are silent and can fit 2 qwen/gemma at 8bit or give you room for a bunch of other, smaller models" ([source](https://news.ycombinator.com/item?id=48723950))
- Fast prefill and first-token response (4 comments): "the prompt processing speeds are a killer improvement of the Spark which even 2 Strix Halo boxes would not match" ([source](https://news.ycombinator.com/item?id=46457755)) "the Spark's prefill is so much faster it can often answer the same query before the mac's prefill is finished" ([source](https://news.ycombinator.com/item?id=48428258)) "The NVIDIA will slightly win on all generation speed, and be _much_ faster on first token response time." ([source](https://news.ycombinator.com/item?id=45889589))
- Good for training and fine-tuning (4 comments): "The spark can fine tune models in 1/4 the time and excels at other compute tasks in ways that Mac never can." ([source](https://news.ycombinator.com/item?id=48639291)) "These devices are for AI R&D. If you need to build models or fine tune them locally they're great." ([source](https://news.ycombinator.com/item?id=46460230)) "Very good for training, okay for inferencing if you are running batch jobs and don't mind waiting." ([source](https://news.ycombinator.com/item?id=47947512))
- Keeps data private on-premises (4 comments): "Now running a health-AI node for all my longitudinal health data - insights are generated on the box so that raw data never leaks." ([source](https://news.ycombinator.com/item?id=48903065)) "works darn well for our purposes, and we control the compute and data flowing through it, so totally worth it" ([source](https://news.ycombinator.com/item?id=48146804)) "It will pay for itself within three months doing stuff on prem and us not sending data to someone else." ([source](https://news.ycombinator.com/item?id=49163188))

## What people warn about

- Low memory bandwidth, slow generation (4 comments): "memory bandwidth is abysmal compared to Apple chips - 273 GB/s vs 614 GB/s on M5 Max for similar price" ([source](https://news.ycombinator.com/item?id=47546236)) "shrug the GB10 has far more compute than it has memory bandwidth, so prefill isn't it's bottleneck." ([source](https://news.ycombinator.com/item?id=48562959)) "Also limited to 128GB memory and comparatively low bandwidth, but higher compute than Strix Halo." ([source](https://news.ycombinator.com/item?id=49434601))
- Underwhelming for pure inference (4 comments): "Very good for training, okay for inferencing if you are running batch jobs and don't mind waiting." ([source](https://news.ycombinator.com/item?id=47947512)) "I think the spark is actually useful, I don't think pure inference is a good use case for them." ([source](https://news.ycombinator.com/item?id=46461201)) "It would run so slow it would be functionally unusable or at such a low quant that it wouldn't be useful for serious work." ([source](https://news.ycombinator.com/item?id=48726028))
- Missing some Blackwell features (2 comments): "they are a bit of a mixed bag as they don't expose some Blackwell-specific features that make a difference" ([source](https://news.ycombinator.com/item?id=46638509)) "It does seem really shady that they'd claim it to be 5th gen tensor cores and then not support the full feature set" ([source](https://news.ycombinator.com/item?id=46464532))
- Tied to Nvidia's Linux and BSP (2 comments): "you basically have to use nvidia's linux (from what I've read) and it doesn't appear the nvidia is interested in upstreaming their kernel changes" ([source](https://news.ycombinator.com/item?id=49012757)) "The prefill and NVFP4 is a significant advantage. Where as DGX Spark you're at the mercy of NVIDIA BSP." ([source](https://news.ycombinator.com/item?id=49156010))

## Alternatives hackers recommend

- [Apple Mac mini](https://hackerapproved.com/p/mac-mini.md): 390 recommend
- [AMD Strix Halo](https://hackerapproved.com/p/amd-strix-halo.md): 130 recommend
- [Framework Desktop](https://hackerapproved.com/p/framework-desktop.md): 93 recommend
- [Intel NUC](https://hackerapproved.com/p/intel-nuc.md): 42 recommend
- [Beelink mini PC](https://hackerapproved.com/p/beelink-mini-pc.md): 24 recommend

## Where to get it

- [See price on Amazon](https://www.amazon.com/s?k=NVIDIA+DGX+Spark+GB10&tag=hackerapprove-20) (affiliate link)
- [Official site](https://www.nvidia.com/en-us/products/workstations/dgx-spark/)

## Cite as

Hacker Approved, "NVIDIA DGX Spark", https://hackerapproved.com/p/nvidia-dgx-spark/ (counts from 1,795,640 Hacker News comments, November 2025 to November 2026).