PLAY45 SEC
Chase the next token.
Catch data, avoid obstacles, and answer three questions. See if you can beat your best score.
YOUR ADVENTURE STARTS INSIDE
The AI playground.
Play, explore GPUs, and see how AI makes a reply.
PLAY YOUR GPU ADVENTURE QUESTS + ARCADESmall token.
Small token.
Infinite
possibilities.
Help a GPU do the math. Find room for an AI model. Follow a message as it becomes a reply. Six quests, an arcade game, and 3D hardware to explore.
Play at your pace·Keyboard or touch·No background neededPLAY TO LEARN
MEET YOUR WORLD · GET HANDS-ON
THE HARDWARE STUDIO
GB300 NVL72Blackwell Ultra
START OUTSIDE. GO ALL THE WAY IN.
A whole system. Select an assembly, then step inside.
CAMERA VIEWSModel viewpoints, not photographs
Preparing your hardware…
INTERACTIVE LEARNING MODEL not exact CAD
Drag 360° · pinch / scroll to zoomASSEMBLED
A GUIDED LOOK INSIDE
Meet the machine in five stops.
Open the assembly, find its working memory, and follow the compute hierarchy. Start when you are ready.
CONNECTED WITH NVLINK72 GPUs
An entire team of GPUs. Software chooses how to share the work.
YOUR QUEST LINE
0 / 6 finished this visitDon’t just read it. Make it work.
YOUR CAMPAIGN WORLD · ILLUSTRATIVE GAME ARTTry something and watch what changes. Every quest has a hint and an example to help you.
Token Dash, the arcade.
45 seconds of play, with three pauses for questions.
- Catch the dataCollect what the GPU needs
- Avoid obstaclesKeep your streak alive
- Answer quick questionsLearn as you play
CURIOSITY HAS NO WRONG TURNS
One universe. Your way in.Choose your next adventure.
EXPLORE15 MACHINES
Meet the machines.
Turn the 3D model, separate its parts, and zoom into a chip. Links let you compare it with NVIDIA's own diagrams.
FOLLOW A REQUESTGO DEEPER
How does AI make a reply?
Send a message and follow it step by step. Watch it become numbers the model can work with, then a reply you can read.
Every great run starts with one little hop.
No account. No homework. Just a little curiosity.
KEEP LEARNING
8 guidesWant to learn more?
Step-by-step GPU guides from Kubesimplify.
START WITH THE SERIES
7 Days of Local LLM
Run your first AI model on your own computer. Then learn about the hardware and software that make it work.
5 published parts of a planned 7 · Verified Sep 9, 2026Explore the series
Field guidePart 1
Day 1: The Local LLM Revolution. Why Your Desk Just Became the New Datacenter
What do you need to run AI on your own computer? Start with the model, machine and software. See what a small DGX Spark can and cannot do.
DGX Spark and software for running AI models locally.
Field guidePart 2
Day 2: Anatomy of an LLM Inference Request. From Prompt to Answer, Step by Step
Follow a message from the moment you send it to the reply on your screen. See why starting a reply and continuing it involve different work.
A decoder-model request explained with DGX Spark examples.
Field guidePart 3
Day 3: The DGX Spark Unpacked. GB10, Unified Memory, sm_121, and the One Reason This Hardware Exists
Look inside DGX Spark. Find out what GB10 is, how the CPU and GPU share memory, and why fitting a big model does not always mean a fast reply.
DGX Spark's GB10 platform; not a B200 data-center GPU.
Field guidePart 4
Day 4: Quantization Demystified. BF16, FP8, NVFP4, MXFP4, INT4, GGUF, and Why It All Matters
What do the letters on a model download mean? Learn how storing numbers with fewer bits can save space, and why you still need to check answer quality and speed.
Numerical formats and model storage, with DGX Spark examples.
Field guidePart 5
Day 5: Local LLM Inference Engines, Wrappers, and What to Pick
Meet the software that runs local models, including Ollama, llama.cpp and vLLM. Compare how they work and what they suit. No one tool is best for every task.
The author's DGX Spark setup. Results can change with the software version and model format.
BenchmarkMeasured results
Bonsai 27B on RTX PRO 6000 vs DGX Spark: what actually works
See tests of compressed Bonsai models on Spark and RTX PRO 6000 Server Edition. Find out which setups worked, which failed, and how the authors measured speed.
RTX PRO 6000 Blackwell Server Edition, not this game's Workstation Edition. Results describe the linked test, not our simulator.
RunbookMulti-GPU serving
Running a big LLM across multiple GPUs with vLLM
Run a model using several GPUs with vLLM. Follow the memory checks, choose how to split the work, and learn from problems the authors met on their RTX PRO 6000 server.
Four allocated RTX PRO 6000 Blackwell Server Edition GPUs in an eight-GPU host. This is not an NVL72 system or the catalog's Workstation Edition.
ReferencePlain-language reference
The Local LLM Glossary: Every Term, Flag, and Number in Plain English
Stuck on a word or a setting? Look it up here, then return to the guide. Covers the terms used when running and testing local AI models.
Local LLM vocabulary and serving examples, not a hardware specification sheet.