Dear NVIDIA, I am extremely disappointed with the current state of DGX Spark. This product was presented as a premium Blackwell platform with FP4, NVFP4, strong local AI capabilities, and a mature software experience. In reality, the experience has felt far less polished and far less complete than…

I guess I knew what to expect after being an early adopter of both Jetson and Shield. Spark is what it is, a hobbyist AI/CUDA platform that will always require expertise and time to keep things working.

I share similar sentiments, but I’m conflicted with placing blame solely on Nvidia. Granted Nvidia did market and is selling the DGX Spark for what it was touted to offer us. Has any of your sentiments been addressed with the Spark’s stakeholders? How much of the DGX Spark is actually built in-hous…

Ubuntu has no blame here. I run it exclusively on all of my machines, even my laptop, and it has been absolutely rock solid stable no matter what I throw at it. An Ubuntu foundation was a big selling point for me with the Spark.

Which Qwen3.5 model did you get 50 t/s? My best output of 122B int4 is about 10+ t/s.

Don’t conflate “blame” with “who is responsible for supporting a given component”. Also, not all NVIDIA employees are Ubuntu users.

[image] Custom built vLLM + Qwen3.5-35B on NVIDIA DGX Spark (GB10) — sustained 50 tok/s, 1M context DGX Spark / GB10 Using this version and it’s lightning fast! Thanks for this! Please post if you do other versions like this. Like the new Qwen versions coming o…

I’m more of a solutions guy myself, we have what we purchased With that said, NVIDIA, the largest company on earth, should see the market opportunity in front of them (AI Agents) as their CEO has, a dedicate a small army of experts ( beside sales people) to solve the issues we all are struggling wit…

Hi everyone, I came across this roadmap in a video from NVIDIA Developer: https://youtu.be/VdIVQcMn0CA?si=P87kJeFUihtLuLZ0 At around the 31:00 mark you can see what’s coming next. Hopefully the upcoming updates will resolve all the issues we’ve been dealing with. [image]

They are on track for the DGX series, but for the Spark, we are stuck in August 2025… [image] See the problem?

Checked on Fedora - the same version looks like I suddenly remembered Torvalds’ words about this brand… dnf provides */nvidia-smi xorg-x11-drv-nvidia-580xx-cuda-3:580.142-1.fc44.x86_64 : CUDA driver tools for xorg-x11-drv-nvidia-580xxRepo : rpmfusion-nonfree UPD: There is such a version…

I am EXTREMely disappointed with the current state of DGX Spark

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

digiegg April 8, 2026, 5:31am 41

check that post for more info. Another solution from non Nvidia but great community members

Topic		Replies	Views
DGX Spark (SM121) Software Support is Severely Lacking - Official Roadmap Needed DGX Spark / GB10	41	5802	February 15, 2026
NVFP4 on DGX Spark / GB10 is broken. I bought 9 of these for this feature. Requesting NVIDIA's official roadmap and response DGX Spark / GB10 jetson , llama , agentic-ai , nemotron , nemoclaw	46	7245	July 17, 2026
Dearest CUTLASS TEAM, When the hell are you going to properly fix tcgen05 FP4 support for DGX Spark / GB10 (SM121)? DGX Spark / GB10	37	2606	April 25, 2026
We unlocked NVFP4 on the DGX Spark: 20% faster than AWQ! DGX Spark / GB10	144	9515	March 14, 2026
FP4 on DGX Spark — Why It Doesn't Scale Like You'd Expect DGX Spark / GB10	213	7279	March 13, 2026
Help: Running NVFP4 model on 2x DGX Spark with vLLM + Ray (multi-node) DGX Spark / GB10 mistral-large	18	2790	December 25, 2025
DGX Spark: 13 → 49 tok/s with Qwen3.5-35B — Native SM121 Kernel Build Guide DGX Spark / GB10 Projects cuda , cusparse	13	1493	April 1, 2026
50%+ Improvement on spark?! DGX Spark / GB10 cuda , deepseek	25	2596	March 24, 2026
DeepSeek-V4-Flash on 4× DGX Spark via vLLM (jasl fork, TP=4, RDMA, MTP) — 49–54 tok/s single-stream, full recipe + the traps DGX Spark / GB10 Projects deepseek	3	817	June 19, 2026
PSA: State of FP4/NVFP4 Support for DGX Spark in VLLM DGX Spark / GB10	234	13873	May 15, 2026

I am EXTREMely disappointed with the current state of DGX Spark

Related topics