Discover
Real workstations with sourced specifications, regional prices and the models known to run on each one.
THE HARDWARE INDEX FOR ON-PREM LLMS
The right hardware for your on-prem AI. Compared with sourced, dated data.
See which AI workstation runs your models, what it costs next to the cloud, and how other teams use it.
Try
THE WORKSTATIONS THE COMMUNITY IS DOCUMENTING
Start from a machine you can point at.
27machines documented
19model configurations
27guides & references
10setups
5deployment records
GB10
DGX Spark
128 GB unified6 model configurations · 5 setups≈ $4.7k Supplier list price · United States · Sept 2026GB300DGX Station
496 GB CPU + 252 GB GPU · 748 GB coherent4 model configurations · 3 setupsPrice from the manufacturer No public list price · manufacturer pageGB10Pro Max with GB10
128 GB unified6 model configurations · 5 setups≈ $8.2k Supplier list price · United States · Sept 2026GB300Pro Max with GB300
496 GB CPU + 252 GB GPU4 model configurations · 3 setups≈ $175k Supplier list price · United States · Sept 2026GB10Ascent GX10
128 GB unified6 model configurations · 5 setups≈ $5.3k Supplier list price · France · LDLC · Sept 2026 · converted from EURGB300ZGX Fury
496 GB CPU + 252 GB GPU4 model configurations · 3 setupsPrice from the manufacturer No public list price · manufacturer pageEdge boards, accelerator chassis and announced prototypes are in the full index.
THIS IS ALREADY HAPPENING
How businesses already use an AI workstation.
01 / THE SERVER ON THE DESK
A company’s AI stack, served by two Sparks in the office.
Understand Tech moved its applications and shared inference from rented cloud GPUs to two GB10 machines on the office floor: retrieval, agents and coding tools for the team.
Founder-reported02 / THE BOX BESIDE THE TEST BENCH
A GB10 beside every test bench in a Wi-Fi NPI lab.
Two agents run the lab’s own test scripts through the controller, judge results against the pass/fail protocols, retest within protocol and report. After a three-month pilot, every bench got its own workstation.
Semiconductor companyOperator report · customer not named
03 / CONFIDENTIAL CLIENT WORK
An accounting firm runs its legal document automation on a GB300, on premises.
“Sending it to the cloud is not acceptable for our clients or for us.” A production GB300 and a development GB10 at the firm; the firm holds the source code of its application.
Accounting & advisory firmCustomer interview · contract scope · firm not named
04 / YOUR DEPLOYMENTCould yours be next?
Named or anonymised, with the hardware, models, applications and what you learned. The Lab team reviews before publication and keeps your name on it.
Add your deploymentWHAT YOU CAN DO HERE
Discover. Learn. Try. Share.
Learn
Use cases from real deployments, operating lessons and recipes that credit their sources and label their evidence.
Plan
Screen a model for memory, plan user load and estimate the full cost per token with clear assumptions.
Share
Add a setup, a failure you solved or a deployment you can talk about. Keep your name on it and follow the machines you run.
START WITH YOUR QUESTION
The questions people bring to the Lab.
Will my model fit in 128 GB?
Memory screening for any model, quantization and context length.
Model fit finderCan one GB10 serve my team?
Plan active users, request rate and response time on a DGX Spark.
Users & workloadWhat does a token cost on a workstation versus cloud?
Hardware, power, software, administration and unused capacity, side by side with an API bill.
Cost calculatorCan agents run locally beside my test equipment?
A Wi-Fi NPI lab where every bench has a GB10 running the execution agent.
Deployment recordWhat still needs the internet?
The privacy boundary, secure operation and the parts of a stack that phone home.
Deployment guideTHE MACHINES ARE NEW. THE CONVERSATION IS OPEN.
Someone else is
figuring this out, too.
Owners, IT leads, developers, integrators and the people buying their first box. A question is a valid first contribution.
Create a profile before 31 December 2026 and it is marked as a founding member of the Lab.
Join the conversationTHE LOCAL AI WATCH
All updates What changed. What it means for your machine.
Loading updates…
WHERE THE LAB STARTED
You see the box. Then you see the possibilities.
We put our own software on GB10 and GB300 machines, spent months making them dependable, and noticed that showing people the actual box changed every conversation. The Lab is that box, opened to everyone.
“Two months ago it was a slide. Today it’s on our floor.”Naama Bak · Co-founder, Understand TechRead the story behind the Lab
Create your free Lab profile.
Save model fits and cost scenarios, follow the machines you run or are considering, and get credited for what you share. Profiles created before 31 December 2026 carry the founding-member mark.
Create a free Lab profile