The typical package uses 2×DGX Spark, two 400G-rated QSFP cables (each Spark port up to 200GbE), Mac mini M4, NAS, and UPS. The listed hardware and accessories total ¥94,700, tax included.
← HomePRIVATE LLM COST ESTIMATOR
Estimate your private LLM budget
Choose a model, concurrency, and business modules to get a tax-included estimate for NVIDIA compute, implementation, and annual operations. No lead form is required.
Your inputs are calculated in this browser and are not stored.SIZING GUIDE
Who is 2×DGX Spark for?
A typical fit is a 20–50-person small or midsize business, or an internal department, with around 10–15 concurrent users. It is especially suitable for midsize law firms and private equity funds that need in-domain document analysis, knowledge retrieval, and internal digital employees. Final sizing still depends on the model build, precision, context, and workload tests.
¥110,000–¥160,000 tax included, covering the mandatory platform, database/backend, model deployment, testing and training modules, plus one ¥20,000 installation visit.
Digital employees, RAG, OCR, DingTalk/WPS integration, and annual operations are calculated separately when needed.
01 · Target model / Peak concurrency
02 · Compute tier
03 · Business modules
The ¥110,000–¥160,000 base service is the sum of the mandatory platform, database/backend, model deployment, testing and training modules. It includes one ¥20,000 installation and on-site commissioning visit, which is not charged again.
Digital employees are priced per business role. RAG pricing covers the vector database only. DingTalk and WPS/Comate are priced in blocks of 50 management accounts.
04 · Annual operations
Annual operations are optional. One workday is calculated as 8 hours.