Nemotron-3-Super-120B: Topology Benchmark on DGX Spark
The model where cluster never loses. Nemotron-3-Super benefits more from TP=2 than any other model tested -- and the SM12.1 CUTLASS patch doubles performance vs FlashInfer.
Archives: March 2026
Found 19 posts
App B The Napkin Math: Predicting Token Speed from Memory Bandwidth
One formula tells you whether your hardware is working correctly or misconfigured. The community used this to diagnose the 40 t/s problem.
App A Why Blackwell's Native MX-Format Support Matters
Blackwell's native MX-format support eliminates the dequantization tax — the hardware and gpt-oss-120b were designed for each other.
GPT-OSS-120B: Topology Benchmark on DGX Spark
Cluster TP=2 dominates decode up to c32, then simple-shuffle takes over. Least-busy routing collapses under load. Full topology comparison with four configurations.
9/9 Claude Code with Local Models: The Series Finale
Everything in the series built toward this: Claude Code running on locally served models. Here's what works, what's rough, and where it's heading.
8/9 The Recipe System: One Command, Zero Flag Archaeology
The recipe system from spark-vllm-docker turns twenty minutes of flag archaeology into one command — and makes everything reproducible.
7/9 Solo vs Cluster: Where Two Sparks Beat One (and Where They Don't)
Neither topology dominates — cluster wins decode at every concurrency level, but 2x Solo wins prefill and TTFT under load. The right choice depends on the workload.
6/9 Two Sparks, One Cluster: Setting Up with Claude Code
The second DGX Spark arrived. Before writing a single line of config: check firmware. Then cables, SSH, Docker, vLLM, model cache — and Claude Code helping build the skills to manage it all.
5/5 From Skills to Agents: What Comes Next
Skills you build today become components for autonomous agents tomorrow. The progression: skill → command → plugin → autonomous agent. Here's where it's all heading.
5/9 LiteLLM: The Translation Layer Between Claude Code and Local Models
Claude Code speaks Anthropic. gpt-oss-120b speaks OpenAI with Harmony-style tool calls. LiteLLM sits in the middle and translates — including a custom callback that patches the tool calls neither side gets right.
4/5 Anatomy of a Plugin: Inside Marp Magic
10 agents. 3 commands. 3 skills. One plugin that turns a topic into a finished presentation. Here's how agents, commands, skills, and hooks work together inside Marp Magic.
4/9 Benchmarking Reality: llama-benchy and the Spark Arena
Post 3 hit 4,158 t/s at c64. llama-benchy puts those numbers under the microscope. Same hardware, two tools, 25x difference in TTFT.
3/9 Switching to vLLM: From 40 t/s to 3,975 t/s
Every reviewer tested single-user latency and called the DGX Spark slow. Nobody tested concurrency. The community found the real number: 3,975 tokens per second.
3/5 Skills That Solve Real Problems: Account Reconciliation
Manual invoice matching: 20% exception rates, hours of CFO time, error-prone. A skill reduces that to under 5% in minutes. Here's how — and it was built by a non-developer.
2/9 First Steps: Running Models on Ollama (20B → 120B)
The standard recipe works but wastes the hardware. Scaling from 20B to 120B on Ollama shows the potential — and the ceiling.
2/5 Your First Skill in 10 Minutes: Meeting Summary
Write a skill file. Run it on messy meeting notes. Get a structured summary. Refine it. Run it again. The whole cycle in 10 minutes — no code, just clear instructions.
1/9 When nvidia-smi Goes Blind: Setting Up btop on DGX Spark
NVIDIA's own monitoring can't see their newest hardware. The community had a fix before NVIDIA did.
1/5 Claude Code Isn't for Developers (It's for Everyone)
You don't need to write code to use Claude Code. Skills are instructions in plain Markdown — if you can write a recipe, you can teach AI to do your repetitive work.
Intro Bring It Home First
Specs can lie in both directions. Snake oil oversells. Reviews undersell. The only truth is your own testing.