AI Servers

EMARQUE's AI Server lineup is the multi-GPU on-prem tier for training, fine-tuning and serving production AI workloads at data-centre scale. The configurations cover the practical range:

• EMARQUE AI Server — 4× / 6× NVIDIA RTX PRO 6000 Blackwell — EMARQUE-built, 6.5U In Win IW-RG650-PRO rackmount, up to 576 GB GDDR7 GPU memory, full PCIe Gen5. Multi-tenant inference, RAG at scale, and fine-tuning models that exceed a single workstation. The 4× starts in the same 6-GPU chassis with room to grow.
• EMARQUE AI Server — 8× NVIDIA RTX PRO 6000 Blackwell — EMARQUE-built, 6.5U In Win IW-RG658-PRO rackmount, 768 GB GDDR7 GPU memory, 9.6kW. The dense-GPU configuration for production inference and larger fine-tunes.
• GIGABYTE G894 NVIDIA HGX B300 AI Server — 8U partner-built HGX B300, 8× B300 Blackwell Ultra with full NVLink and 2.1 TB HBM3E, your choice of AMD EPYC or Intel Xeon. Frontier LLM training and high-throughput inference, with configuration flexibility and lower TCO.
• NVIDIA DGX B300 AI Server — NVIDIA's own 10U turnkey system, 8× B300 Blackwell Ultra, 2.1 TB HBM3E, 144 PFLOPS sparse FP4, the complete DGX software stack and NVIDIA Enterprise Services. The foundation of NVIDIA DGX SuperPOD.
• NVIDIA Vera Rubin AI Server — next-generation Vera Rubin platform. Specifications and Malaysia availability confirmed on enquiry as NVIDIA finalises the launch.

All are pre-order via EMARQUE solution providers. Final configuration, lead time and pricing confirmed on enquiry.

EMARQUE AI Servers fit teams running AI as production infrastructure, not exploration. Common profiles: enterprise data-science and ML teams deploying internal LLMs, AI-product engineering teams serving customer-facing inference at scale, AI research labs training mid-to-frontier models on-prem, university HPC programmes, AI service providers and AI-foundry-style internal platforms, sovereign-AI deployments where data cannot leave the country. The common thread: workloads that exceed a single DGX Station or workstation, the need for predictable cost and latency, and a requirement that the stack stay on-premise.

EMARQUE AI Server with RTX PRO 6000 Blackwell — 96 GB GDDR7 per GPU, PCIe Gen5, designed for inference density and mixed AI plus compute. Four-, six- and eight-GPU configurations in EMARQUE-built 6.5U In Win rackmount chassis. Strong for multi-tenant serving, RAG, embeddings, fine-tuning, video and vision workloads. The most cost-effective entry into multi-GPU on-prem AI, and the most flexible — same GPU, scale the count to your workload.

B300 Blackwell Ultra tier — 8× B300 GPUs on the NVLink baseboard, 2.1 TB HBM3E memory pool, for frontier LLM training and the highest-throughput inference. Two ways to buy it: the GIGABYTE G894 (8U partner-built HGX B300, AMD or Intel CPU, configuration flexibility and lower TCO), or the NVIDIA DGX B300 (NVIDIA's own 10U turnkey system with the full DGX software stack, Mission Control and NVIDIA Enterprise Services, the foundation of DGX SuperPOD). The step up when full NVLink GPU-to-GPU bandwidth and HBM3E are the constraint.

NVIDIA Vera Rubin — the next-generation Vera Rubin architecture (Vera CPU + Rubin GPU). Sits above Blackwell Ultra. Available to order; full specifications and Malaysia availability confirmed on enquiry as NVIDIA's launch finalises.

All three use the same NVIDIA RTX PRO 6000 Blackwell GPU (96 GB GDDR7 each) — you scale the GPU count to your workload.

4× — 384 GB aggregate GDDR7, in the 6.5U In Win IW-RG650-PRO chassis (4 of 6 bays populated, expandable to 6). Sized for multi-tenant inference for a small team, RAG and embeddings at department scale, fine-tuning up to mid-tier model sizes. The entry into multi-GPU on-prem AI, with headroom to grow.

6× — 576 GB aggregate GDDR7, the IW-RG650-PRO fully populated. The balanced production box — higher inference concurrency and larger fine-tunes without moving to the larger chassis.

8× — 768 GB aggregate GDDR7, in the larger 6.5U In Win IW-RG658-PRO chassis (9.6kW). The densest RTX PRO 6000 box — whole-organisation inference serving, the largest fine-tunes, maximum GPU density per box. Requires a datacentre-grade 3-phase rack PDU.

Practical rule: 4× to start with room to grow, 6× for balanced production, 8× for maximum single-box density. EMARQUE AI sizes any of them against your real workload (email business@emarque.co).

NVIDIA DGX Station is a desk-side AI workstation built on the GB300 Grace Blackwell Ultra Superchip — 748 GB coherent memory, 20 PFLOPS FP4, MIG up to 7 users, 1,600W. Ideal as a powerful single-user or small-team AI workstation that lives in an office.

EMARQUE AI Servers are rackmount platforms designed for production AI in a data centre or server room: multi-GPU density (4–8 GPUs per box, in 6.5U In Win chassis for the RTX PRO 6000 Blackwell tier, 8U for the GIGABYTE G894 HGX B300, and 10U for the NVIDIA DGX B300), datacentre power and cooling, designed to compose into clusters, run as production infrastructure rather than as a desk workstation.

Choose DGX Station when one box, in an office, served to a small team via MIG is enough. Choose an AI Server when production load, multi-tenant inference, or training at scale exceeds a single Station — or when the deployment belongs in a proper rack with datacentre conditions.

NVIDIA DGX Spark is the personal AI tier: GB10 Grace Blackwell, 128 GB unified memory, ~1 PFLOPS FP4, sits on a desk. Designed for AI development, prototyping, RAG and fine-tuning small-to-mid LLMs.

EMARQUE AI Servers are the production tier: rack-scale multi-GPU systems running frontier-model training and high-throughput inference. The progression is Spark for development → Station for serious single-user / small-team on-prem AI → AI Server for production multi-GPU infrastructure. EMARQUE helps size the right tier against your actual workload.

NVIDIA HGX B300 — shipping now. Production-ready frontier inference and training. Known specifications, known availability through NVIDIA OEM partners. The right choice if you need to deploy this year.

NVIDIA Vera Rubin — next-generation architecture, available to order with delivery aligned to NVIDIA's launch timeline. Specifications and Malaysia availability confirmed on enquiry. The right choice if your deployment timeline aligns with the platform's launch and you want the latest-generation hardware.

EMARQUE AI helps align your deployment timeline with the right NVIDIA generation. Email business@emarque.co.

AI Servers are datacentre-class equipment. Indicative figures (exact per-unit specifications confirmed on enquiry):

• Power: RTX PRO 6000 4× configurations typically 2–3 kW per server; 8× configurations 3–5 kW. The B300 systems (GIGABYTE G894, NVIDIA DGX B300) run higher — redundant 80 PLUS Titanium PSUs and 3-phase rack power are typical. Confirm with EMARQUE AI before specifying your rack PDU.
• Cooling: all current configurations are air-cooled (direct-liquid options available on the B300 platforms). Hot-aisle / cold-aisle datacentre conditions, front-to-back airflow, sufficient CFM at the rack — standard datacentre requirements.
• Rack: standard 19-inch — 6.5U for the RTX PRO 6000 In Win servers, 8U for the GIGABYTE G894, 10U for the NVIDIA DGX B300. Deep rails for full PSU and cable clearance.
• Networking: typically 800 Gb/s ConnectX-8 SuperNIC plus 400 Gb/s for compute fabric, plus management 10 GbE.

EMARQUE AI handles deployment scoping, rack and PDU sizing, and integration with your existing datacentre as solutions on request.

For multi-server training and tightly-coupled inference, the network fabric matters as much as the GPUs. AI Server configurations support NVIDIA ConnectX-8 SuperNICs at up to 800 Gb/s Ethernet or InfiniBand, plus QSFP112 ports at 400 Gb/s for compute fabric. HGX B300 uses NVIDIA's full NVLink interconnect inside the server (GPU-to-GPU at NVLink bandwidth). Rack-scale platforms additionally use NVLink Switch for full-bandwidth GPU-to-GPU across the rack.

EMARQUE AI handles NCCL and MPI tuning, ConnectX-8 / QSFP112 deployment, rack-level fabric design, and integration with your existing datacentre network as solutions on request.

The full NVIDIA AI Enterprise stack — CUDA, container runtimes, NIM microservices for production inference, AI Workbench, NVIDIA Base Command for multi-tenant orchestration, and the full library of pre-built framework images (PyTorch, vLLM, RAPIDS, NeMo). EMARQUE-built RTX PRO 6000 servers ship with Ubuntu LTS plus the NVIDIA AI Enterprise stack; NVIDIA HGX B300 ships with NVIDIA DGX OS plus the full NVIDIA AI Enterprise + Base Command stack as the reference configuration.

EMARQUE AI solutions on request can layer in a ready-to-use private LLM, RAG, embeddings, and agentic-AI stack pre-configured for your data and use case — chat UI, vector search, document ingestion, model serving, all on-prem.

EMARQUE is among the first solution providers in Malaysia accepting pre-orders for the NVIDIA AI Server lineup and EMARQUE-built RTX PRO 6000 configurations. Pre-ordering with EMARQUE:

• Locks your allocation for delivery in Malaysia.
• Final pricing confirmed on enquiry — pricing varies by configuration (GPU count, networking, storage, software stack). Listed prices are indicative estimates only.
• Direct access to the EMARQUE AI team during pre-order for deployment scoping — rack, networking, fabric, software stack, team training.
• Delivery window confirmed at order, aligned with the OEM partner's Malaysia shipping schedule.

Click ENQUIRE NOW on any variant page or email business@emarque.co.

AI Server pricing is configuration-driven and not a fixed retail price. Final pricing depends on GPU count and SKU, system memory, NVMe storage, ConnectX-8 / QSFP112 networking configuration, software stack and licensing, deployment scope, and EMARQUE AI integration services selected. Listed indicative prices anchor expectations only; final quotes are produced on enquiry against your specific configuration and rack requirements. The EMARQUE AI team can scope and quote a full deployment, or just supply the hardware — whichever fits.

The EMARQUE AI Server (4×, 6× and 8× RTX PRO 6000 Blackwell configurations) is an EMARQUE own-brand platform built in a 6.5U In Win rackmount chassis — PSU, motherboard, NVMe, networking and cooling are integrated by EMARQUE in Malaysia, with NVIDIA RTX PRO 6000 Blackwell GPUs inside. EMARQUE handles assembly, validation, stress-testing, BIOS and driver tuning for AI workloads, and Malaysia-based warranty + RMA. The GIGABYTE G894 (HGX B300) is a GIGABYTE partner-built server; the NVIDIA DGX B300 and Vera Rubin are NVIDIA's own platforms — EMARQUE handles solution delivery and local support for all of them.

EMARQUE is an AI server solution provider and systems integrator in Malaysia — not just a reseller. The EMARQUE AI team handles sizing and configuration; rack and PDU specification; ConnectX-8 / QSFP112 networking deployment; multi-server cluster setup; Base Command and Kubernetes orchestration; private LLM / VLM / RAG / agent stack deployment; team training; and Malaysia-based post-sale support — all available as EMARQUE AI Solutions on request. Scope, timing and charges quoted on enquiry. Email business@emarque.co.

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.