Somehow related: "The end of my AArch64 [Ampere Altra Q80-30] desktop experiment", https://news.ycombinator.com/item?id=48728599 / https://marcin.juszkiewicz.com.pl/2026/06/26/the-end-of-the-...

I strongly considered it, but the one thing that scares me away from wanting to do the spark is 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.

I'm avoiding where possible buying electronics where support is controlled by the manufacturer and not me.

I do appreciate how Nvidia tries to say close to vanilla with Linux and Android (nvidia shield). Instead of trying to build a shitty moat like Samsung with all their garbage software.

If nvidia ever releases Android smartphone, I'd probably stand in line to get one.

Is Nvidia better than Intel

"Multi-token prediction gives a free speedup of up to 2x on many models" - at the expense of halving prompt processing speed

This is an interesting review. I have a Chinese strix halo box that's isnt available in the west (favm faex1) I've been able to do some ok graphical gen, or some decent agentic tasks as a fallback for when some of the APIs are overloaded during business hours, but nothing amazing for sure, and also not both at the same time. But here's the thing...it cost me 1800usd two months ago....and it's runs x86. I am struggling to see why people pay +2x more for the Arm Nvidia version, despite the slightly higher bandwidth it still does basically the same AI tasks and alot fewer high end general computing tasks... I like my box but I wouldn't find it useful enough to pay more than I did for it or get more of them and cluster for instance. Can anyone explain the allure of the Nvidia box, other than brand name?

I have been running uConsoles with CM5 (2712 and 3588 with 16GB RAM) for 6 months as daily drivers.

They are ~$500* and present the same ARM problems/opportunities.

But they are completely silent (no fan, the case is the heat sink).

My 6600(3050) desktop from 2016(2024) with replaced SSD(2021)/RAM(2025) (they age like milk) now gets little use and M$ will soon sleep with the fishes.

*Hard to get now as the 3588 that has linux for uConsole is out of stock and the Raspberry one is rare and more expensive by the day.

Neat blog! I was intrigued by this bullet point mentioned in passing:

>my four hard drive USB 3.2 ZFS raidz2 array with four 24 TB drives

Can you speak more about this? Which USB array did you choose? How well does it work? I've been slowly planning a transition away from my power-hungry surplus enterprise gear in the 19" rack towards a smaller, quieter, lower power setup ... but storage is the real kicker right now. I have a 12x18TB array in raidz2 built into a 1U NAS case, and I just can't quite figure out a better way to package something like that. I would need three USB arrays if I want to reuse the existing drives, which I think I do given how expensive storage is today.

It's interesting how many of these issues don't appear to be specific to the DGX Spark but to the standard "Nvidia GPUs suck on Linux" type of issues that afflict a lot of people.

Super-powerful (if rather pricy) Linux desktop that happens to play games while doing everything else... that man is living my dream :'D

Please don’t buy a DGX Spark unless all three of these are true:

  - You value simplicity more than performance or price-to-performance.
  - You accept that the hardware will depreciate rapidly.
  - You’re prepared to buy two or four of them.

OR:

  - You want to run frontier models right now as cheaply as possible
  - You want to run high-parameter models on a 15a breaker/line

Otherwise, get a normal, high-bandwidth GPU.

A single Spark gives you roughly 115 GB of usable memory compared with the 24–32 GB found on many lower-cost GPUs. It's certainly a big increase, but in practice it does not unlock dramatically better models.

  - One Spark: More memory, but mostly enough for poor-quality, extremely low-bit quants of larger models.  
  - Two Sparks: Enough for mid-tier parameter models at reasonable quants, such as DeepSeek V4 Flash and HY3.
  - Four Sparks: Enough for GLM 5.2 at a reasonable quant.  You'll need a $1000+ switch too.

The problem is that Sparks are slow compared with almost everything else in their price range. Many factors affect inference speed, but memory bandwidth is one of the biggest. A $4,000-plus DGX Spark provides only 273 GB/s.

  | GPU                 | Memory bandwidth |           VRAM | Approx. price  |
  | ------------------- | ---------------: | -------------: | ------------:  |
  | DGX Spark           |         273 GB/s | ~115 GB usable |        $4,000+ |
  | RTX 5060            |         448 GB/s |          16 GB |          $600  |
  | Radeon AI Pro R9700 |         640 GB/s |          32 GB |        $1,200  |
  | RTX 4000 Pro        |         672 GB/s |          24 GB |        $2,300  |
  | RTX 4500 Pro        |         896 GB/s |          32 GB |        $3,500  |
  | RTX 3090            |         936 GB/s |          24 GB |        $1,200  |
  | RTX 5000 Pro        |       1,344 GB/s |          48 GB |        $6,000  |
  | RTX 5090            |       1,792 GB/s |          32 GB |        $4,000  |
  | RTX 6000 Pro        |       1,792 GB/s |          96 GB |       $12,000  |

Yes, the Spark has substantially more memory. But going from roughly 24 GB to 115 GB does not necessarily unlock substantially better model quality. In many cases, it only lets you load heavily compressed 2-bit versions of larger models, such as DeepSeek V4 Flash, with serious quality degradation.

24–32 GB is currently a sweet spot. Models such as Qwen 3.6 27B and 35B-A3B:

  - Perform far above what their parameter counts suggest.
  - Fit comfortably within 24–32 GB of VRAM at reasonable quantization levels.

A 4-bit quant of Qwen 3.6 27b (18 GB) will out-perform a 2-bit quant of DeepSeek v4 Flash (90gb).

Instead of the Spark, if I had a roughly $4,000 budget...

Assuming I already had a reasonably modern desktop:

  - One RTX 5090, RTX 5000 Pro, or RTX 4500 Pro.
  - Two RTX 3090s, RTX 4000 Pros, or R9700s, provided the motherboard can bifurcate two physical x16 slots into x8/x8.

If I were building a system from scratch:

  - A DDR4- or PCIe 4.0-era consumer CPU and motherboard that supports x8/x8 bifurcation.
  - Two RTX 3090s, RTX 4000 Pros, or R9700s.

If I were already planning to buy a new Mac:

  - A MacBook Pro M5 with 64 GB or 128 GB of unified memory.

For context, these are the systems I currently run:

  - EPYC Turin with four RTX 6000 Pro Max-Qs.
  - EPYC Milan with four RTX 3090s.
  - AM4 with two RTX 3090s.
  - AM4 with two RTX 3090s.
  - Intel Raptor Lake with two RTX 5060 Ti.
  - MacBook Pro M3 128GB Unified

I'm genuinely disappointed with my Spark. I don't know how anyone can claim it performs decently with LLMs or diffusion models. Back when I worked in VFX in the early 2000s, we had a saying: "Render time is coffee time" and if you try to run this thing with a usable context size, you'll be drinking a lot of coffee. Most of the optimizations it relies on for inference simply aren't available for training, so it crawls like a snail on almost every model. An RTX 6000 Blackwell would have been the better investment for an AI enthusiasts and for general computing there are cheaper offerings.

Yes. Waiting for the Windows Version myseflf. RTX Spark Desktop.

This is something I'd do if I've had the disposable income lol

>Non-Steam games have a lower chance of working

Wonder if it's true for GOG games because they are usually installed in a neatly packaged folder without any bloat.

I would avoid the DGX Spark. For that price and its performance on running local models it is a complete scam. This tweet says it all [0]

[0] https://xcancel.com/petergostev/status/1978230978725507108

It's simple, the 395+ Max Strix Halo you bought for $1800 is now a ~$4000 build (at least the AMD AI dev unit). If only we had time travel right? Either way, the Nvidia unit comes with Connect-X 7. That may or may not matter to you, but the hardware for that isn't cheap. In general the Nvidia cards also have better support for models. I know AMD is trying to catch up but anything except their datacenter cards do not seem to be getting a lot of attention.

> isnt available in the west [...] Can anyone explain the allure of the Nvidia box

The first part might answer the second one. Otherwise, the lack of CUDA and the nvidia ecosystem of tooling could also explain why it doesn't seem so interesting for AI tasks.

I'm actually confused by the post - indeed why choosing the 27B dense and the 35B MoE models? Given the structure of the post I honestly have the impression that the Spark was purchased out of curiosity more than for a specific purpose (LLMs).