NOREX

Local AI Workstation

Shipped in a state where AI already runs

Power it on and the local LLM server starts. No setup, no API billing, no usage caps. Your data never leaves the machine.

  • 24GB VRAM, keeps a 30B-class model resident
  • 32,768 token context, four concurrent users
  • No API billing, works offline

Market structure

Nothing sits between 2.7M and 8M KRW

Entry-level brand workstations ship with 4GB VRAM or integrated graphics, so a 30B-class model will not load. Machines that actually run local LLMs start around 8 million KRW.

  1. 230~270만원Entry brand models (RTX A400 4GB or integrated graphics)Cannot run
  2. 399만원DELL Pro Max Tower T2 (integrated graphics, 8GB memory)Cannot run
  3. 400~450만원NX-24 (RTX 3090 24GB)Runs 30B class
  4. 800~960만원128GB unified memory classRuns 30B class
  5. 1,123만원 이상Workstations with RTX 5090 or aboveRuns 30B class

Based on a survey of 1,364 workstation listings on Danawa

Products

Two configurations

NX-24 Base

메모리 32GB

Preparing for sale

로컬에서 30B급 모델을 상주 구동하는 AI 워크스테이션

  • 24GB VRAM, 30B급 모델을 4bit 양자화 상태로 상주 구동
  • 컨텍스트 32,768 토큰
  • 동시 4명 사용 가능

4,000,000KRW

View details

NX-24 Pro

메모리 64GB

Preparing for sale

대용량 시스템 메모리로 여유를 확보한 상위 구성

  • 24GB VRAM, 30B급 모델을 4bit 양자화 상태로 상주 구동
  • 컨텍스트 32,768 토큰
  • 동시 4명 사용 가능

4,500,000KRW

View details

Why this configuration

24GB VRAM

This is the threshold where a 30B-class model fits entirely in 4bit quantized form. A 4GB card cannot load the model at all.

936 GB/s memory bandwidth

Token generation speed scales almost linearly with memory bandwidth. This is the most bandwidth available in this price range.

Local-only configuration

The OpenAI-compatible API server binds to localhost only. Data stays on the machine, with no request caps or metered billing.

Inspection report included

Only parts that pass every inspection step are fitted, and the measurements are supplied as a document.

Inspection and warranty

Only parts that pass inspection go in

Every incoming part goes through an inspection process. Parts that do not pass never enter a product, and the measurements are supplied as a document.

Every incoming part inspected
Graphics cards go through sustained load and temperature measurement. Storage devices are checked for prior usage.
One year warranty on the whole product
One year from delivery, two years for the mainboard. Parts are held for four years from manufacture.
Inspection report and warranty certificate included
You can verify the exact part models and inspection measurements in writing.

See full specifications

Preinstalled

It arrives already set up

Operating system, drivers, inference server and model files are installed and configured before shipping. Power on and the local API server starts automatically.

  • Windows 11
  • NVIDIA 그래픽 드라이버
  • llama.cpp (로컬 LLM 추론 서버)
  • Gemma 4 31B 모델 파일 (17.53GB)
  • 로컬 API 서비스 (포트 8090, 로컬 전용, 외부 접속 불가)
  • GPU 전력 관리 설정 (280W 제한)

Questions about configuration or lead time

Tell us what you plan to run and we will help you pick the right configuration.

Contact us