DGX Spark
From RM 15,888
- Best for
- On-prem AI development, prototyping, and evaluation
- Form factor
- Compact desktop
- GPU memory pool
- 128 GB unified (CPU + GPU coherent)
- Configurable?
- No — NVIDIA OEM SKU
- Status
- Shipping now
Compare
Decision matrices across the lineup.
Three named decision guides, plus per-family compare tables. Pick the right system class, GPU generation, and refresh window without bouncing between data sheets.
Decision guides
Pick a question, get a structured answer.
Workstation vs Server vs Rack-Scale
Pick the right class before drilling into specs. Sized by users, model class, deployment shape, and budget.
Open guide
H200 vs B200 vs B300
The three NVIDIA GPUs you'll be quoted in 2026. HBM, FP4, NVLink generation, and which EMARQUE systems carry each.
Open guide
Generations: Hopper → Rubin
Hopper, Blackwell, Blackwell Ultra, Rubin — what's shipping, what's next, and how to plan a multi-year refresh.
Open guide
Personal AI & Small Scale
2 systems in this class — side-by-side.
Desk-side and small-scale systems for developers, individual researchers, and small teams.
AI Work 100
From RM 12,000
- Best for
- Individual developers and researchers running 7B–70B class models locally
- Form factor
- Tower workstation, standard 240 V wall outlet
- GPU memory pool
- Up to 192 GB DDR5-6400 (4 DIMM)
- Configurable?
- Yes — EMARQUE-built to order
- Status
- Shipping now
AI Workstation
2 systems in this class — side-by-side.
Pedestal multi-GPU workstations for individual to departmental AI — assembled and supported locally by EMARQUE.
DGX Station
—
- Best for
- Training and fine-tuning models up to 1 trillion parameters locally
- Form factor
- —
- GPU memory pool
- 748 GB coherent (252 GB HBM3e + 496 GB LPDDR5X)
- Configurable?
- No — NVIDIA OEM SKU
- Status
- —
PRO 500
From RM 35,000
- Best for
- Departmental private RAG with 25-100 concurrent users
- Form factor
- —
- GPU memory pool
- Up to 1 TB ECC DDR5-6400 (8 channels)
- Configurable?
- Yes — EMARQUE-built to order
- Status
- Shipping now
AI Server
3 systems in this class — side-by-side.
Single-node rackmount AI servers — EMARQUE AI Server (RTX PRO 6000 SE) and NVIDIA DGX B200 / B300 with HGX OEM alternatives.
AI Server
—
- Best for
- Departmental private RAG / chat — 100–500 concurrent users on 70B-class models
- Form factor
- 4U rackmount, 19" rails included
- GPU memory pool
- Up to 2 TB ECC DDR5-6400 (24 channels, dual-socket)
- Configurable?
- Yes — EMARQUE-built to order
- Status
- —
DGX B200
Contact for pricing
- Best for
- Generative AI training and inference
- Form factor
- 10U rackmount
- GPU memory pool
- 1.4 TB HBM3e (180 GB × 8)
- Configurable?
- No — NVIDIA OEM SKU
- Status
- Shipping now
DGX B300
Contact for pricing
- Best for
- AI reasoning model training and inference
- Form factor
- 10U rackmount, air-cooled
- GPU memory pool
- Up to 2.3 TB HBM3e per node (288 GB × 8)
- Configurable?
- No — NVIDIA OEM SKU
- Status
- Pre-order / allocation
AI Factory
2 systems in this class — side-by-side.
Rack-scale NVLink-Switch fabrics for foundation-model training and sovereign compute — GB300 NVL72, DGX GB300, and the Vera Rubin generation.
GB300 NVL72
Contact for pricing
- Best for
- Trillion-parameter foundation model training
- Form factor
- Single rack — building block for DGX SuperPOD
- GPU memory pool
- Up to ~21 TB HBM3e per rack (288 GB × 72)
- Configurable?
- No — NVIDIA OEM SKU
- Status
- Shipping now
Vera Rubin NVL72
Contact for pricing
- Best for
- Multi-year AI Factory and sovereign-compute roadmaps
- Form factor
- —
- GPU memory pool
- HBM4 (capacity per GPU per NVIDIA's GA datasheet)
- Configurable?
- No — NVIDIA OEM SKU
- Status
- Roadmap
Contact Us
Get in Touch with Us
Tell us about your workload. We reply within one business day with a quote sized to fit.
- 01
Key Account Manager
+6012 627 2280 - 02
Request for Quotation
business@emarque.co