8× NVIDIA B200 Tensor Core GPUs
Fifth-generation NVLink and two NVSwitch chips interconnect all eight GPUs into a coherent 1,440 GB HBM3e memory pool. 8 TB/s per-GPU memory bandwidth.
Shipping nowNVIDIA DGX
NVIDIA DGX B200
Foundation for the AI Center of Excellence — eight NVIDIA B200 Tensor Core GPUs in a unified, air-cooled DGX system.
Pricing on request — allocation and configuration confirmed at quotation.
8B200 GPUs
1,440 GBHBM3e
144PFLOPS FP4 inference
Key features
Configuration overview.
Manufacturer-defined features from the published datasheet.
72 PFLOPS training · 144 PFLOPS inference
FP4 precision performance per system, as published by NVIDIA. Tensor Cores accelerate FP4, FP6, FP8, FP16, BF16, TF32, and FP32 workloads.
Dual Intel Xeon Platinum 8570
112 cores total at 2.1 GHz base / 4.0 GHz max boost. 4 TB of DDR5 system memory. NVIDIA-validated CPU and memory configuration.
8× ConnectX-7 + 2× BlueField-3 DPU
Per-GPU 400 Gb/s InfiniBand or 200 GbE scale-out networking via NVIDIA ConnectX-7 VPI. Two NVIDIA BlueField-3 DPUs offload storage and infrastructure services.
NVIDIA DGX OS + NVIDIA AI Enterprise
Pre-installed software stack: NVIDIA DGX OS, NVIDIA AI Enterprise (CUDA, cuDNN, NCCL, TensorRT-LLM, NeMo, NIM microservices), and NVIDIA Base Command for cluster orchestration.
Three-year NVIDIA Enterprise Support
NVIDIA Business Standard Support hardware coverage and NVIDIA AI Enterprise software entitlement included as standard with every NVIDIA DGX B200 system.
Two ways to buy
Same platform — choose the supply path.
The NVIDIA HGX baseboard is identical on both paths. The turnkey DGX is fastest to deploy; OEM HGX platforms (Dell, Giga Computing, Supermicro) give wider configuration choice.
NVIDIA
NVIDIA DGX B200 (turnkey)
NVIDIA-built and NVIDIA-supported reference platform. Ships pre-configured with NVIDIA DGX OS, NVIDIA AI Enterprise, NVIDIA Base Command, and a three-year NVIDIA Enterprise Support contract.
- 10U air-cooled NVIDIA DGX reference chassis
- NVIDIA DGX OS + NVIDIA AI Enterprise + NVIDIA Base Command
- Three-year NVIDIA Enterprise Support contract included
- Reference building block for NVIDIA DGX SuperPOD with B200 systems
Dell · Giga Computing · Supermicro
HGX B200 — OEM platforms
Same NVIDIA HGX B200 baseboard from NVIDIA's OEM partners. Customer-selectable CPU, memory, storage, and networking configurations within each OEM's published configuration matrix. OEM warranty and support model. NVIDIA AI Enterprise software available separately.
- Dell PowerEdge XE9680 / XE9685L — air-cooled HGX B200, dual Xeon
- Giga Computing G893-SD1 / G893-ZX1 — 8U HGX B200, dual Xeon or EPYC
- Supermicro SYS-821GE-TNHR / AS-A126GS-TNBR — HGX B200 (Intel / AMD)
- Customer-selectable CPU, memory capacity, NVMe topology, networking
EMARQUE supplies both paths in Malaysia. Final configuration, lead time, and warranty terms are confirmed in writing at quotation.
Architecture
Under the hood.
The four sub-systems that determine real-workload behaviour. We tune each before delivery.
GPU complex
- 8 × NVIDIA B200 SXM (HGX B200 reference baseboard)
- 1,440 GB HBM3e total (180 GB × 8) · 8 TB/s per-GPU bandwidth
- 5th-gen NVLink + 4× NVSwitch · 1.8 TB/s per GPU all-to-all
- FP4 / FP6 / FP8 inference acceleration · 72 PFLOPS dense FP4 training
CPU & system memory
- Dual Intel Xeon Platinum 8570 (56 cores, Emerald Rapids)
- Up to 4 TB DDR5 ECC memory across the host platform
- PCIe Gen5 to HGX baseboard
- NVIDIA-validated CPU + memory configuration (no other CPU options)
Scale-out networking
- 8 × NVIDIA ConnectX-7 — 400 Gb/s InfiniBand or 200 GbE per GPU
- GPUDirect RDMA over InfiniBand for multi-node tensor parallelism
- Quantum-X InfiniBand or Spectrum-X Ethernet fabrics supported
- Optional: in-band management dual-port 100 GbE for orchestration
Power, cooling, form factor
- 10U rackmount chassis (NVIDIA DGX B200 reference)
- ~14.3 kW peak — 6 × 3.3 kW 80+ Titanium PSUs (3+3 redundant)
- Air-cooled — inlet temperature ≤ 27 °C recommended
- DGX OS · NVIDIA AI Enterprise · CUDA · NeMo · NIM · TensorRT-LLM
Next step
Get a DGX B200 configuration and lead time from your Malaysian NVIDIA systems specialist.
Supported workloads
Reference workload categories.
Workload categories documented in the manufacturer's reference materials. Sizing is confirmed with your technical team during scoping.
Generative AI
Foundation model training and inference
Train and deploy large language models, multi-modal models, and reasoning models on a single coherent 1,440 GB HBM3e memory pool. Tensor Core FP4 acceleration supports models with extended context windows.
AI development
Unified develop-to-deploy pipeline
NVIDIA DGX B200 is positioned by NVIDIA as a single platform spanning the AI development lifecycle — data preparation, fine-tuning, evaluation, and production inference deployment via NVIDIA NIM microservices.
DGX SuperPOD
Reference building block for cluster scale-out
Eight NVIDIA ConnectX-7 ports per system provide NVIDIA Quantum-2 InfiniBand connectivity for multi-node NVIDIA DGX SuperPOD configurations. Reference architectures published by NVIDIA cover deployments from 32 to 1,024+ nodes.
Scientific computing
HPC, simulation, and data analytics
Accelerate scientific simulation, computational biology, climate modelling, and large-scale data analytics workloads with NVIDIA CUDA-X libraries on the eight-B200 NVLink-coherent compute platform.
Full spec sheet
Every line documented at quotation.
As supplied by NVIDIA. EMARQUE handles in-country delivery, commissioning, and Tier-1 support handoff.
- GPU
- 8 × NVIDIA B200 Tensor Core (NVLink 5)
- GPU memory
- 1.4 TB HBM3e (180 GB × 8)
- FP4 compute
- 72 PFLOPS training · 144 PFLOPS inference (per node)
- CPU
- Dual Intel Xeon Platinum 8570 (56-core)
- System memory
- 4 TB DDR5
- Networking
- 8 × NVIDIA ConnectX-7 400 Gb/s InfiniBand / 200 GbE
- Storage
- 2 × 1.9 TB OS NVMe · 8 × 3.84 TB data NVMe (30 TB raw)
- Power
- 14.3 kW max (six 3.3 kW PSUs, 3+3 redundant)
- Cooling
- Air-cooled
- Form factor
- 10U rackmount
- Software
- NVIDIA DGX OS · AI Enterprise · CUDA · NIM · NeMo
FAQ
Common questions about DGX B200
What is included with NVIDIA DGX B200?
Per NVIDIA's published datasheet: eight NVIDIA B200 Tensor Core GPUs interconnected via fifth-generation NVLink and two NVSwitch chips, dual Intel Xeon Platinum 8570 CPUs (112 cores total), 4 TB DDR5 system memory, 30 TB internal NVMe storage (2× 1.9 TB OS + 8× 3.84 TB data), 8× NVIDIA ConnectX-7 VPI cards, 2× NVIDIA BlueField-3 DPUs, NVIDIA DGX OS, NVIDIA AI Enterprise software entitlement, NVIDIA Base Command, and a three-year NVIDIA Business Standard Support hardware contract. 10U air-cooled rackmount chassis.
What are the published performance figures?
Per NVIDIA: 72 petaFLOPS of FP4 training performance per system, 144 petaFLOPS of FP4 inference performance per system, 1,440 GB total HBM3e GPU memory (180 GB × 8) with 8 TB/s per-GPU bandwidth. Maximum system power 14.3 kW.
How does NVIDIA DGX B200 differ from HGX B200 OEM systems?
Both use the same NVIDIA HGX B200 baseboard with eight B200 SXM GPUs. NVIDIA DGX B200 ships in NVIDIA's reference 10U chassis with NVIDIA DGX OS, NVIDIA AI Enterprise software, NVIDIA Base Command, and a three-year NVIDIA Enterprise Support contract bundled. HGX B200 OEM systems from Dell (PowerEdge XE9680 / XE9685L), Giga Computing (G893-SD1 / G893-ZX1), and Supermicro (SYS-821GE-TNHR / AS-A126GS-TNBR) use the same HGX baseboard but the OEM's chassis design, customer-selectable CPU / memory / storage configurations, and the OEM's warranty and support model. NVIDIA AI Enterprise software is licensed separately for OEM systems.
What is the form factor and site requirement?
10U rackmount chassis. 14.3 kW maximum system power served by six 3.3 kW Titanium-rated PSUs in a 3+3 redundant configuration (200–240 V AC, 50/60 Hz). Air-cooled. Operating temperature range 5–30 °C per NVIDIA's published specifications.
Is NVIDIA DGX B200 a building block for NVIDIA DGX SuperPOD?
Yes. NVIDIA DGX B200 is the reference building block for NVIDIA DGX SuperPOD with B200 systems. Multi-node configurations connect via NVIDIA Quantum-2 InfiniBand at 400 Gb/s per GPU. NVIDIA publishes reference architectures for SuperPOD configurations from 32 nodes upward.
Can systems from different vendors be mixed in the same cluster?
Yes. The NVIDIA HGX B200 baseboard is the same reference design used in NVIDIA DGX B200 and in OEM HGX B200 systems from Dell, Giga Computing, and Supermicro. Systems from different vendors interoperate at the InfiniBand fabric layer and run the same NVIDIA AI Enterprise software stack.
What is the typical lead time?
Lead time follows NVIDIA's allocation schedule for NVIDIA DGX B200. EMARQUE confirms projected delivery window at order acknowledgement following NVIDIA allocation. Multi-node NVIDIA DGX SuperPOD projects are scoped individually.
Request configuration & quotation.
Manufacturer specifications, factory lead times, and warranty terms apply. EMARQUE responds within one business day with a formal quotation and projected delivery window.
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