강문규

@devsnack

Nvidia dgx sparks gb10, Asus gx10, llm, local ai

Seoul, South KoreaJoined May 2026

About

Software developer from South Korea.

Available for

Nothing here yet.

강문규's blogs

DevSnackdevsnack.hashnode.dev3 posts

Articles Comments

Recently published

강강문규devsnack.hashnode.devMay 6 · 10 min read

Qwen3.6 on DGX Spark: vLLM + NVFP4 + DFlash vs llama.cpp — 2x Faster at 88–104 tok/s

TL;DR — I was happily running Qwen3.6 on llama.cpp. Then I saw claims of 2× speed with vLLM + NVFP4 + DFlash. So I installed it, fought through crashes, and measured it myself. Verdict: it's real. 88–

0

강강문규devsnack.hashnode.devMay 6 · 8 min read

Gemma 4 MTP Drafter on DGX Spark: 2.89x Speedup for Dense 31B — No Quality Loss

An 870 MB drafter model turned Dense 31B from 6.5 → 18.8 tok/s. No model swap, no training, no quality degradation. If you have a DGX Spark, there's no reason not to use this. Key Results Model Fra

0

강강문규devsnack.hashnode.devMay 6 · 8 min read

Every Optimized Model for NVIDIA DGX Spark GB10 — Benchmarked & Ranked (April 2026)

Got your hands on an NVIDIA DGX Spark but have no idea which models to run on it? I scoured every GB10-optimized model on Hugging Face so you don't have to. Table of Contents What Makes DGX Spark Spe

1

L

강문규

About

Available for

강문규's blogs

Recently published

Qwen3.6 on DGX Spark: vLLM + NVFP4 + DFlash vs llama.cpp — 2x Faster at 88–104 tok/s

Gemma 4 MTP Drafter on DGX Spark: 2.89x Speedup for Dense 31B — No Quality Loss

Every Optimized Model for NVIDIA DGX Spark GB10 — Benchmarked & Ranked (April 2026)

Search Hashnode

강문규

About

Available for

강문규's blogs

Recently published

Qwen3.6 on DGX Spark: vLLM + NVFP4 + DFlash vs llama.cpp — 2x Faster at 88–104 tok/s

Gemma 4 MTP Drafter on DGX Spark: 2.89x Speedup for Dense 31B — No Quality Loss

Every Optimized Model for NVIDIA DGX Spark GB10 — Benchmarked & Ranked (April 2026)