MSI XpertStation WS300 NVIDIA DGX Station Grace Blackwell Ultra Computer

The MSI XpertStation WS300 is MSI's NVIDIA DGX Station partner unit — a desktop AI supercomputer built on NVIDIA's GB300 Grace Blackwell Ultra Desktop Superchip reference design. Each unit ships as a complete system: the GB300-based tower, its internal 1,600W ATX power supply, MSI's documentation, and NVIDIA's DGX OS preinstalled. For teams deploying serious on-prem AI, EMARQUE AI Solutions are available on request — pre-validation testing, deployment scoping, MIG partitioning, ConnectX-8 networking integration, private LLM/RAG/agent stack setup.

MSI's XpertStation line carries the brand's enterprise-workstation pedigree — refined chassis engineering, qualified for sustained AI workloads. MSI is one of NVIDIA's launch partners for the DGX Station family alongside ASUS, Dell, HP, Supermicro, and Gigabyte; all units share the GB300 silicon, so the meaningful differences between partner variants come down to chassis design, thermal envelope, storage layout, and I/O integration.

The MSI XpertStation WS300 follows NVIDIA's published DGX Station GB300 reference specification:

• Superchip: NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip — Grace 72-core Arm Neoverse V2 CPU + Blackwell Ultra GPU, linked via NVLink-C2C at 900 GB/s.
• Total coherent memory: 748 GB — 252 GB HBM3e GPU memory @ 7.1 TB/s + 496 GB LPDDR5X CPU memory @ 396 GB/s, unified across CPU and GPU.
• AI compute: up to 20 PFLOPS FP4 (sparse; 15 PFLOPS dense); 10 PFLOPS FP8/FP6; 5 PFLOPS FP16/BF16; 2.5 PFLOPS TF32; 1.3 TFLOPS FP64.
• Model ceiling: supports models up to 1 trillion parameters.
• Networking: NVIDIA ConnectX-8 SuperNIC at up to 800 Gb/s; 2× QSFP112 @ 400 Gb/s each; 1× RJ45 10 GbE.
• Storage: 2× M.2 2280 PCIe 5.0 x4 NVMe (from the CPU) + 2× M.2 2280 PCIe 6.0 x4 NVMe for training data. Drive capacities are configuration-dependent — confirmed at order.
• MIG: partitions the GPU into up to 7 isolated instances for multi-user serving.
• Power & cooling: 1,600W 80 PLUS Titanium ATX PSU; direct liquid cooling for CPU + GPU with dual 360 mm radiators.
• OS: Ubuntu with the NVIDIA AI software stack and CUDA-X libraries, preinstalled and configured for the system. NVIDIA AI Enterprise is a separate licence, not bundled with the hardware — EMARQUE can include a term in your quote.

Chassis: 245 × 528.4 × 595 mm (W × H × D) tower. Expansion: 1× double-wide PCIe 5.0 x16 slot (the slot a display GPU occupies) + 2× single-wide PCIe 5.0 x16 slots running at x8 signal + 1× PCIe x1 M.2 for WiFi/Bluetooth. Out-of-band management via ASPEED AST2600 (IPMI 2.0, Redfish) on a dedicated 1 GbE port. Email business@emarque.co for final configuration.

NVIDIA DGX Station is a desktop AI supercomputer built on the GB300 Grace Blackwell Ultra Superchip, designed for serious on-prem AI workloads — training and fine-tuning trillion-parameter models, multi-user inference (MIG up to 7 instances), agentic AI with NemoClaw, RAG at scale, vision-language models, physical AI, and massive-dataset data science via RAPIDS. Target users: AI research labs, enterprise data science teams, university HPC programmes, AI-product engineering teams. The step up from DGX Spark when one Spark is too small and workstation-class GPUs aren't enough.

Warranty terms for the MSI XpertStation WS300 are confirmed at order — we state them on the quote rather than in general copy, so what you are told matches what you get. EMARQUE’s role is Malaysia-based local handling:

• Local point of contact — pre-sale, post-sale and warranty all routed through the same EMARQUE team. A real person, not a ticket queue.
• RMA routing — if a hardware fault appears, we handle the manufacturer claim on your behalf rather than leaving you to navigate the process.

For deeper post-sale needs — pre-validation testing, OS reinstall, driver issues, MIG configuration, ConnectX-8 networking, multi-Station setup, private AI stack deployment — solutions are available from the EMARQUE AI team on request. Scope and charges quoted on enquiry.

Same family, very different scale. DGX Spark: GB10 Grace Blackwell, 128 GB unified memory, ~1 PFLOPS FP4, models up to ~200B (practical ~120B), 240W — the personal AI development unit. DGX Station: GB300 Grace Blackwell Ultra, 748 GB coherent memory, 20 PFLOPS FP4 sparse (15 PFLOPS dense), models up to 1 trillion parameters, MIG up to 7 users, 1,600W.

Spark = local AI development, prototyping, fine-tuning small-to-mid LLMs. Station = serious on-prem AI infrastructure, multi-user inference, frontier-scale training. EMARQUE AI helps you size the right tier (email business@emarque.co).

LLM training and fine-tuning up to 1T-parameter scale — pre-training, SFT, RLHF, DPO, LoRA / QLoRA at production sizes. Multi-user inference via MIG up to 7 isolated partitions — one Station serves a whole team. Agentic AI with NVIDIA NemoClaw for secure tool-calling and autonomous workflows. Vision-language models, multimodal, physical-AI training for robotics/autonomous systems. Massive-dataset data science via RAPIDS. Real-world: private GPT-class fine-tuning on enterprise data, multi-team inference, AI Foundry-style internal LLM deployments.

The system arrives ready to work: Ubuntu with the NVIDIA AI software stack, preinstalled with system-specific configuration and the essential CUDA-X libraries, plus GPU and networking drivers. You can pull framework containers (PyTorch, vLLM, RAPIDS, NeMo) and start work without building the stack yourself.

NVIDIA AI Enterprise is licensed separately. It is not bundled with the hardware — it adds the supported, production-grade tier: NIM microservices, enterprise support and security maintenance on the frameworks. If you need it, EMARQUE quotes the licence term alongside the system so it arrives together rather than as a later surprise.

NVIDIA NemoClaw, announced at GTC in March 2026, is NVIDIA’s open-source platform for governed agentic AI — runtime controls, sandboxing and policy guardrails for always-on agents. It runs on this hardware and is a common reason teams choose a Station over a workstation GPU.

EMARQUE AI Solutions on request can layer in a ready-to-use private LLM, RAG and agent stack pre-configured for your data and use case.

StorageReview’s independent review (24 August 2026) named the XpertStation WS300 its Best GB300 System for local AI. Measured results from that review:

• GPT-OSS-120B: up to 9,294 output tokens/sec — roughly 2.7× an RTX PRO 6000 Blackwell on the same workload.
• GPT-OSS-20B: 22,161 tokens/sec at 128 concurrent streams — versus ~1,469 on a DGX Spark.
• MiniMax M2.7 (125 GB, NVFP4): 193 tokens/sec single-stream; 4,801 tokens/sec at peak concurrency.
• Frontier-scale checkpoints, locally: ran GLM-5.2 (433 GB) and Nemotron-3-Ultra 550B with CPU-memory offload — model sizes no single-GPU workstation can load.
• Thermals: the liquid-cooled GB300 peaked at 71°C under a sustained 1,292W load — desk-quiet, not server-hall.

Trade-offs the same review notes honestly: the Arm-based Grace CPU excludes some x86-only toolchains, and there is no display output without an add-in GPU (BMC video only) — worth confirming your stack in the pre-order conversation.

The MSI XpertStation WS300 is on pre-order now, with units arriving in Malaysia at the end of September 2026. Pre-ordering with EMARQUE:

• Secures your unit from the end-September arrival — stock is limited.
• Pricing confirmed on enquiry — the quote covers configuration options, the NVIDIA AI Enterprise term and any EMARQUE AI services scope.
• Complimentary MSI MAG 272QP QD-OLED monitor (26.5" WQHD, 99% DCI-P3, worth RM 2,299) with the first 10 WS300 pre-orders — while the promotional allocation lasts.
• Direct access to the EMARQUE AI team during pre-order for deployment scoping — MIG partitioning, ConnectX-8 networking, private LLM stack, team training.

Click ENQUIRE NOW above, email business@emarque.co or WhatsApp 012 627 2280 to pre-order.

EMARQUE is a Malaysian solution provider for on-prem AI. What EMARQUE brings around a DGX Station deployment is the depth of solutions available from the EMARQUE AI team:

• Deployment solutions on request — MIG partitioning, ConnectX-8 / QSFP112 networking setup, NemoClaw integration for agentic AI, private LLM / RAG / agent stack deployment, team training. Scope and charges quoted on enquiry.
• Workflow sizing consultation — DGX Station vs DGX Spark vs RTX PRO 6000 vs AI server. Honest assessment of which tier fits your real workload.
• Pre-validation testing on request — bench-test the unit for thermal performance and stability before dispatch.
• Local point of contact — pre-sale sizing, post-sale technical questions, warranty handling — same Malaysian team, no offshore queue.
• EMARQUE AI stack + assisted setup — pre-configured local stack (chat UI, RAG with vector search, document ingestion, model serving) installed and tuned for your data.

For buyers comparing solution providers on the same hardware, the practical question is: who actually knows this product in country and can help when you need it? Email business@emarque.co for sizing.

Yes. EMARQUE AI is our dedicated AI compute division — full NVIDIA AI hardware stack and on-prem deployment for Malaysian businesses, research labs, and enterprises: NVIDIA Personal AI (DGX Spark, 2× Spark stacking), AI workstations (RTX PRO 6000 Blackwell and others), DGX Station GB300, multi-GPU AI servers, plus full integration support. Visit EMARQUE.AI or contact business@emarque.co.