Skip to content
EMARQUE.AI
  1. Home
  2. Products
  3. NVIDIA DGX Station
Pre-order · Launching 2026

Trillion-parameter AI,
desk-side.

NVIDIA's GB300 Grace Blackwell Ultra Superchip in a workstation chassis. 748 GB of coherent memory, 20 petaFLOPS of FP4 compute, and ConnectX-8 networking — enough capacity to train and run trillion-parameter models locally. Three OEM variants. EMARQUE handles pre-order, delivery, setup, and warranty in Malaysia.

Pre-Order with EMARQUESee variants on emarque.co
●NVIDIA-Powered3 OEM VariantsRM 0Local Pre-Order
NVIDIA DGX Station with GB300 Grace Blackwell Ultra — desk-side AI supercomputer supplied by EMARQUE in Malaysia
20PFLOPS FP4 sustained153 PFLOPS sparse peak. GB300 Grace Blackwell Ultra Superchip on the desk.
748GB coherent memory252 GB HBM3e on the Blackwell Ultra GPU + 496 GB LPDDR5X on the Grace CPU.
800Gb/s ConnectX-8 networkingPlus 2 × 400 GbE QSFP112 + 1 × 10 GbE — multi-node training fabric in the chassis.
Why DGX Station

A trillion-parameter capable system, on the desk.

GB300 brings memory and compute densities that previously meant a rack-scale deployment to a single workstation chassis — with office-grade power and acoustics.

Trillion-parameter local AI

748 GB of coherent memory and 20 petaFLOPS of FP4 compute let you train, fine-tune, and run models up to 1 trillion parameters without leaving your premises.

HBM3e + LPDDR5X unified

252 GB of HBM3e on the GPU + 496 GB of LPDDR5X on the Grace CPU, linked by NVLink-C2C at 900 GB/s. No host-device data shuffle.

Multi-user inside one box

Up to 7 isolated MIG partitions per GB300 — give individual researchers or services their own slice of the accelerator with hardware-enforced isolation.

Full DGX software stack

Ships with DGX OS + NVIDIA AI Enterprise. CUDA, NIM, AI Workbench, NeMo, vLLM, RAPIDS, and PyTorch images all pre-configured for the GB300 platform.

Desk-side, not server-room

Workstation form factor with workstation acoustics. 1,600 W on standard single-phase AC. No data centre upgrade, no cold-aisle planning.

EMARQUE local pre-order + setup

Lock your allocation now in MYR. EMARQUE coordinates delivery with NVIDIA's OEM partners, then handles site readiness, network onboarding, and warranty fulfilment locally.

OEM variants

Three OEM variants, one platform price.

All three variants ship the same GB300 Grace Blackwell Ultra Superchip, 748 GB coherent memory, ConnectX-8 networking, and DGX software stack — priced at RM 499,999 each. Differences are chassis design, serviceability, and ISV validation focus.

Workstation pedigree

MSI XpertStation WS300

RM 0
Pre-order · MYR

MSI's XpertStation enterprise workstation line. Industrial reliability, generous service access, validated for ISV workflows.

Pre-order on emarque.co
Quiet enterprise

ASUS ExpertCenter Pro ET900N G3

RM 0
Pre-order · MYR

Part of ASUS's ExpertCenter Pro line. Premium materials, thermal headroom for sustained loads, designed for office or lab placement.

Pre-order on emarque.co
Service-friendly

Gigabyte W775-V10-L01

RM 0
Pre-order · MYR

Gigabyte's GB300 workstation. Tool-less serviceability, validated airflow path, generous front-panel I/O.

Pre-order on emarque.co
EMARQUE includes with every DGX Station pre-order
  • Pre-order allocation locked at order time with EMARQUE
  • Site readiness review (power, cooling, network drop)
  • On-site delivery + unboxing + asset registration
  • Network onboarding (IP plan, ConnectX-8 fabric configuration)
  • Stack validation — runs your first model and produces a benchmark report
  • Onboarding workshop for your team (DGX OS, MIG, AI Workbench)
  • Local warranty fulfilment via the OEM channel — coordinated by EMARQUE
Architecture

What's inside the GB300 platform.

NVIDIA defines the platform; OEMs deliver the chassis. Below is the common compute, memory, networking, and software stack you get with any of the three OEM variants.

Compute

  • NVIDIA GB300 Grace Blackwell Ultra Superchip
  • Grace CPU — 72-core Arm Neoverse V2
  • Blackwell Ultra GPU — next-gen Tensor Cores with FP4 and FP8 acceleration
  • Up to 20 PFLOPS FP4 sustained · 153 PFLOPS sparse peak
  • NVLink-C2C 900 GB/s CPU ↔ GPU coherent interconnect

Memory

  • 252 GB HBM3e (Blackwell Ultra GPU side)
  • 496 GB LPDDR5X (Grace CPU side)
  • 748 GB total coherent address space
  • No PCIe bottleneck — single memory model across CPU + GPU

Networking

  • NVIDIA ConnectX-8 SuperNIC — 800 Gb/s Ethernet
  • 2 × QSFP112 — 400 Gb/s each
  • 1 × RJ45 10 GbE for management
  • Multi-station clustering supported for distributed training

Software & Power

  • DGX OS (Ubuntu-based) + NVIDIA AI Enterprise
  • CUDA · NIM · AI Workbench · NeMo · vLLM · RAPIDS · PyTorch images
  • Up to 7 MIG isolated partitions per GB300
  • 1,600 W single-phase AC — no dedicated circuit, no data centre cooling
Use cases

What DGX Station unlocks.

Frontier AI training

Training and fine-tuning trillion-parameter models

GB300's combined HBM3e + LPDDR5X address space holds frontier-class models in memory. LoRA, QLoRA, and full fine-tunes on Llama 4 / DeepSeek-class models locally.

Enterprise data science

Heavy data-science and research teams

Multiple researchers share one chassis via MIG partitions. JupyterHub-style multi-tenancy with hardware-enforced isolation per user or per project.

Agentic AI

Production agentic AI inference

Long-context agents, planner-executor loops, and multi-step tool use need the FP4 throughput and unified memory the GB300 platform delivers.

Pre-cluster

Single-box validation before multi-node DGX cluster

Validate your stack and your model recipes on one DGX Station before scaling to a multi-node DGX SuperPOD. Same software, same runtime, same OS.

Is this for you?

Client guidance — DGX Station.

How EMARQUE scopes this system: who it suits, when to pick it, when to pick something else, and what we add beyond the hardware.

Who it's for

Research labs, data-science teams, and serious solo practitioners running NVIDIA's desk-side flagship for local Blackwell Ultra-class work.

Choose this when

  • You need NVIDIA DGX Station as an NVIDIA-branded desk-side system with the full DGX OS / AI Enterprise stack.
  • Your workload benefits from 748 GB coherent memory in a single desk-side chassis (up to 7 MIG instances).
  • Pre-production validation for workloads that will eventually scale to a DGX SuperPOD.

Pick something else when

  • Workload runs comfortably on Spark or AI PRO 500 — Station is significantly more expensive.
  • Real growth path is rack-scale — go to DGX B200 / B300 or GB300 NVL72 directly.
  • Acoustic or single-phase power constraints make a desk-side 1.6 kW system impractical.

Best-fit workloads

  • Training and fine-tuning models up to 1 trillion parameters locally
  • Enterprise data science teams and research labs running heavy workloads
  • Agentic AI and multi-user inference inside a single chassis (up to 7 MIG instances)
  • Pre-production validation before scaling to multi-node DGX clusters

Model & memory fit

Up to trillion-parameter models locally per NVIDIA's published figures. 252 GB HBM3e + 496 GB LPDDR5X coherent memory holds large KV caches and long context windows resident.

Deployment shape

Desk-side tower; 1.6 kW single-phase AC, conditioned air recommended. Includes ConnectX-8 800 GbE for upstream cluster joins.

Alongside the rest of the lineup

NVIDIA's desk-side flagship — three OEM variants (MSI XpertStation WS300, ASUS ExpertCenter Pro ET900N G3, Gigabyte W775-V10-L01). Step-up from DGX Spark when 128 GB unified memory is no longer sufficient; step-down from rack-class DGX systems when desk-side fits the deployment.

Upgrade path

When desk-side performance is exceeded, move to single-node DGX B200 / B300 (10U rackmount) or rack-scale GB300 NVL72.

What EMARQUE adds beyond the hardware

  • NVIDIA-powered systems specialist in Malaysia — local MYR pricing.
  • OEM variant selection guidance (MSI / ASUS / Gigabyte) based on the customer's software stack.
  • Pre-delivery installation of customer-selected runtime against the validated DGX OS image.
  • Local NVIDIA Enterprise Support escalation path.
Next step

Get a DGX Station configuration and lead time from your Malaysian NVIDIA systems specialist.

Talk to a DGX Station specialistCompare systemsContact us
Platform specs

Full specification window.

Superchip
NVIDIA GB300 Grace Blackwell Ultra
CPU
Grace 72-core Arm Neoverse V2
Interconnect
NVLink-C2C 900 GB/s (CPU ↔ GPU)
Memory
748 GB coherent (252 GB HBM3e + 496 GB LPDDR5X)
AI compute
Up to 20 PFLOPS FP4 · 153 PFLOPS sparse peak
Networking
ConnectX-8 SuperNIC 800 GbE · 2 × QSFP112 400 GbE · 1 × 10 GbE
MIG partitions
Up to 7 isolated instances
Software
DGX OS · AI Enterprise · CUDA · NIM · AI Workbench · PyTorch · vLLM · RAPIDS · NeMo
Power
1,600 W (single-phase AC)
OEM variants
MSI XpertStation WS300 · ASUS ExpertCenter Pro ET900N G3 · Gigabyte W775-V10-L01
FAQ

DGX Station in Malaysia — common questions.

  • What is the price of NVIDIA DGX Station in Malaysia?
    All three OEM variants — MSI XpertStation WS300, ASUS ExpertCenter Pro ET900N G3, and Gigabyte W775-V10-L01 — are priced at RM 499,999 each through EMARQUE. Pricing is in MYR and includes EMARQUE's local pre-order, delivery coordination, setup, and warranty handoff.
  • When does NVIDIA DGX Station ship?
    DGX Station is officially launching in 2026 via NVIDIA's OEM partners. EMARQUE is taking pre-orders now — placing a pre-order with us locks your allocation when partner units arrive. We update pre-order customers monthly on channel timing.
  • Which DGX Station OEM variant should I choose?
    All three variants share the same GB300 Grace Blackwell Ultra Superchip, 748 GB coherent memory, and DGX software stack. Differences are chassis design, serviceability, and ISV validation focus. MSI XpertStation has the longest enterprise workstation pedigree; ASUS ExpertCenter Pro leads on premium materials and office-friendly acoustics; Gigabyte W775 emphasises service-friendliness and airflow path. We'll match you to one at quotation.
  • Can DGX Station train trillion-parameter models locally?
    Yes. The 748 GB combined coherent memory (252 GB HBM3e + 496 GB LPDDR5X) supports model sizes up to ~1 trillion parameters with appropriate quantization. For full-precision frontier training a multi-station cluster is recommended; the on-board ConnectX-8 800 GbE networking makes that linkage straightforward.
  • Does DGX Station need a data centre or special power?
    No data centre is required. The chassis runs on standard single-phase AC at 1,600 W — typically a dedicated 13 A circuit is sufficient. Office-grade air conditioning handles the thermal load. EMARQUE provides a site readiness checklist with every pre-order.
  • How does DGX Station compare to DGX Spark?
    DGX Spark is the entry-level personal AI supercomputer (GB10 Grace Blackwell, 128 GB unified memory, RM 15,888 – RM 24,999) for individual developers and edge deployments. DGX Station is the desk-side flagship (GB300 Grace Blackwell Ultra, 748 GB coherent memory, RM 499,999) for trillion-parameter training and multi-user inference. Same software stack, completely different scale.
  • What does EMARQUE include with the DGX Station?
    Pre-order allocation, site readiness review, delivery, unboxing, network onboarding (including ConnectX-8 fabric setup), stack validation with benchmark report, an onboarding workshop for your team, and local warranty fulfilment via the OEM channel. See the inclusion list above for the full scope.

Pre-order your DGX Station.

NVIDIA's OEM partners launch in 2026 — pre-ordering with EMARQUE now locks your allocation, sets site readiness, and gets you on the local warranty channel from day one.

Pre-Order with EMARQUE+6012 627 2280

Related searches · NVIDIA DGX Station Malaysia · DGX Station GB300 Malaysia · Grace Blackwell Ultra Malaysia · trillion parameter workstation · desk-side AI supercomputer Malaysia · MSI XpertStation WS300 · ASUS ExpertCenter Pro ET900N G3 · Gigabyte W775-V10-L01 · HBM3e workstation Malaysia · ConnectX-8 800 GbE workstation · DGX Station pre-order Malaysia

Contact Us

Get in Touch with Us

Tell us about your workload. We reply within one business day with a quote sized to fit.

Build your workstationContact us
  1. 01

    Key Account Manager

    +6012 627 2280
  2. 02

    Request for Quotation

    business@emarque.co