We’ve just open-sourced Atlas, an LLM inference engine purpose-built (but not limited to) GB10-class hardware, Repository is here: https://github.com/Avarok-Cybersecurity/atlas and we need the communities help to keep elevating it for developers. Many DGX Spark owners and Blackwell developers r…

Qwen3.5-122B-A10B-NVFP4 on a single DGX Spark just verified its working. [image] whpthomas: Same error – this doesn’t run either Run command: sudo docker run -d --name atlas-122b-test \ --gpus all --ipc=host --network host \ …

How many medium article ‘breakthroughs’ have you had since then jumping from attempts to ‘fix’ vllm by completely breaking it and misunderstanding simple cuda architecture flags to atlas? Edit: this isn’t a rhetorical question, added quotes to breakthrough and legitimately would like a number

Regarding Qwen 3.6 27B, that 13 tok/sec number is without MTP. With MTP you can expect 20 tok/sec on the Spark. Not great, but it’s a free boost. This depends on the prompt of course. @AzeezIsh Congrats on the release, I hope your project grows and improves. My advice based on reading this thread i…

Has anyone tested Qwen3.6 27B? What’s the speed like? I use a custom llama.cpp fork with dflash and get around 20-40, depending on the task. VLLM with MTP3 gets around 30-35.

[image] tbraun96: We can certainly improve the README and hygiene of keeping the docs synced to the codebase going forward. Maybe a AI assistant would be helpful ? Seriously though, I have not tested this, so no comments on the performance but seeing the initial feedbacks here, I hesitate to …

I do agree, in any case the effort spent by the developers of atlas should be respected. On the other hand, what I did not like here is the fact the Atlas team tend to present this solution as revolutionary but as a matter of fact and as for now, it does not seem to perform any acceptable realistic …

How are the talks with nvidia and qwen going? I’ll just leave this here for people that don’t think you’re lying. [image] We unlocked NVFP4 on the DGX Spark: 20% faster than AWQ! We are working with Nvidia to include this in their official community image. [image] Introducing the Atlas Infe…

Maybe it’s because I’m more used to the self-aggrandizing typical in the ‘startup posting on social media’ era and am better at reading between the lines, but you’re acting in really bad faith by calling it lies and it’s becoming annoying. I actually tried it last night and they delivered improvemen…

The criticisms of the tool itself are already well expressed in this thread. There’s no ad hominem attack in play, just highlighting the deception used in the past by the author regarding the apparent validity of projects. The example you responded to never amounted to anything and Nvidia never ‘i…

[image] trystan1: ‘Being accustomed to exaggerated claims by startups’ is not a valid argument and a scary precedent to set. Sadly, the “fake it till you make it” strategy does work. I’m against it, hate it, but I have to admit that in my 40 years career, I know of many that would argue that …

On Sparkrun, for Qwen 3.6’s 35B MoE, we shatter records. Official Nvidia and AMD accounts have retweeted several of our posts, as well as donated physical hardware to help bring local inferencing to everybody. The hostility in this forum comes from those who have nothing to show, and waste their hum…

Atlas: Open-source inference engine for DGX Spark <2minute cold start, 100+ tok/s on Qwen3.6-35B-FP8, 13+ supported models

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

trystan1 May 9, 2026, 10:33am 80

How are the talks with nvidia and qwen going? I’ll just leave this here for people that don’t think you’re lying.

Edit: adding the claim that ’ no vibe code is in place ’

Topic		Replies	Views
Introducing the Atlas Inference Server and Engine DGX Spark / GB10 Projects	163	11913	June 30, 2026
We unlocked NVFP4 on the DGX Spark: 20% faster than AWQ! DGX Spark / GB10	144	11346	March 14, 2026
Athena's Engine: DeepSeek and Qwen3.8 Next — 1,126 tokens/s prefill, 32 tokens/s decode, 256k ctx on 1DGX Spark DGX Spark / GB10 Projects deepseek	11	1287	September 23, 2026
Running nvidia/Qwen3.8-Flash-Next-NVFP4 on a single DGX Spark DGX Spark / GB10	14	2302	September 19, 2026
Qwen/Qwen3.6-35B-A3B (and FP8) has landed DGX Spark / GB10 agentic-ai	309	34533	June 22, 2026
veloGB10 v0.4.0 RELEASE: Qwen 3.8-27b code sustained tok/s: 4x : 125 tok/s , 2x: 85/tok/s, 1x: 75tok/s DGX Spark / GB10 Projects	117	6082	September 26, 2026
Benchmark Report: Qwen3.6-35B-A3B-NVFP4 on NVIDIA DGX Spark, Jetson Thor, Blackwell 6000 Pro DGX Spark / GB10 Projects	12	6115	August 21, 2026
Qwen3.5-397B-A17B + DGX Spark (duo) DGX Spark / GB10 Projects	62	7429	June 14, 2026
DGX Spark + Qwen3-Next-80B: Proven Performance, But Missing Clear Path to NIM, TensorRT-LLM & Web UIs DGX Spark / GB10 cuda , nim , llama	16	5937	March 6, 2026
Qwen3.8-Flash-Next recipe for 1x GB10, At mtp=3 c=1, code is ~47.5t/s, json is ~60t/s. Prefix cache OK. Intel AutoRound int4 quant DGX Spark / GB10 llama	12	2475	September 9, 2026

Atlas: Open-source inference engine for DGX Spark <2minute cold start, 100+ tok/s on Qwen3.6-35B-FP8, 13+ supported models

Related topics