Eight Blackwell Ultra GPUs
Eight NVIDIA B300 Blackwell Ultra SXM GPUs provide 288 GB of HBM3e memory per GPU, giving 2,304 GB installed memory, which NVIDIA lists as 2.3 TB total.
NVIDIA DGX B300
$240,000.00
NVIDIA DGX B300 is a 10U enterprise AI system built for large-scale generative AI, reasoning-model inference, model training and accelerated data-center workloads. It integrates 8× NVIDIA B300 Blackwell Ultra SXM GPUs with 8 × 288 GB HBM3e GPU memory (2,304 GB installed; approximately 2.3 TB total), dual Intel Xeon Platinum 6776P processors, fifth-generation NVLink/NVSwitch connectivity, NVIDIA ConnectX-8 networking and BlueField-3 DPUs.
Organizations looking to buy NVIDIA DGX B300 or compare DGX B300 price, availability and deployment requirements can request a destination-specific B2B quotation. The quoted system should define the exact AC/PDU or DC/busbar power configuration, NVIDIA support scope, freight and professional installation requirements.
Explore Related
Related products, categories and research.
AI Servers & Infrastructure
NVIDIA DGX B300 AI system
NVIDIA DGX B300 is an enterprise AI supercomputing system for organizations building and operating modern AI factories. The 10U platform combines eight NVIDIA Blackwell Ultra SXM GPUs, dual Intel Xeon processors, high-capacity system memory, NVMe storage, NVLink/NVSwitch fabric, NVIDIA ConnectX-8 networking and BlueField-3 DPUs in a validated DGX architecture.
It is designed for demanding workloads such as large language model inference, reasoning models, generative AI training, retrieval-augmented generation, multimodal AI, fine-tuning and high-throughput enterprise AI services.
8× B300Blackwell Ultra SXM GPUs
2.3 TB8 × 288 GB HBM3e
144 PFLOPSFP4 Tensor Core performance
Product overview
NVIDIA DGX B300 for enterprise AI infrastructure
DGX B300 is a complete NVIDIA DGX system, not a PCIe workstation graphics card. NVIDIA positions it as a foundation for AI reasoning and large-scale enterprise AI deployment. The system is built around 8× NVIDIA B300 Blackwell Ultra GPUs and provides the compute density, memory capacity, networking and software integration required for advanced AI training and inference.
The platform is compatible with modern data-center rack deployments and is offered with AC/PDU and DC/busbar power approaches. This makes infrastructure planning—rack capacity, power delivery, airflow, networking and facility readiness—an important part of procurement.
For buyers comparing NVIDIA DGX B300 price, availability or configuration, the final commercial quotation should identify the exact power version, support terms, delivery scope and any deployment services required for the destination data center.
Three checks before purchase
1
Confirm the power version
Identify whether the deployment requires the AC/PDU configuration or DC/busbar architecture and confirm facility power capacity.
2
Validate rack and network readiness
DGX B300 is a 10U system with high-density power, cooling and up to 800 Gb/s cluster-network connectivity.
3
Confirm support and delivery scope
Document the DGX support term, included hardware, freight, installation responsibilities and commissioning requirements in the quotation.
Technical highlights
NVIDIA DGX B300 key capabilities
The DGX B300 is engineered as an integrated AI system rather than a collection of standalone accelerator cards.
High-throughput AI performance
NVIDIA lists up to 144 PFLOPS of FP4 Tensor Core performance and 72 PFLOPS of FP8 Tensor Core performance for DGX B300.
NVLink system fabric
Two NVIDIA NVLink Switch systems provide high-bandwidth GPU-to-GPU communication, with NVIDIA listing 14.4 TB/s aggregate NVLink bandwidth.
Dual Intel Xeon processors
The system uses two Intel Xeon Platinum 6776P processors and is configured with 2 TB of system memory by default, with support for up to 4 TB.
800 Gb/s cluster networking
Eight OSFP connections serve eight NVIDIA ConnectX-8 VPI adapters, supporting up to 800 Gb/s NVIDIA InfiniBand or Ethernet connectivity.
Integrated enterprise storage
Internal storage includes eight 3.84 TB E1.S NVMe self-encrypting drives plus two approximately 1.9 TB M.2 NVMe drives for the operating system.
Applications and fit
What NVIDIA DGX B300 is designed for
DGX B300 targets compute-intensive enterprise AI workloads where memory capacity, accelerator interconnect, high-speed networking and validated system software are critical.
LLM inference and reasoning
Run large language models, reasoning models and high-throughput inference services that benefit from Blackwell Ultra compute and large GPU memory.
Generative AI training
Train and fine-tune large AI models using an integrated eight-GPU platform with high-bandwidth NVLink communication.
RAG and enterprise AI
Support retrieval-augmented generation, multimodal workflows, enterprise copilots and data-intensive AI applications.
Scale-out AI factories
Deploy DGX B300 as part of larger NVIDIA DGX BasePOD or SuperPOD architectures using high-speed InfiniBand or Ethernet fabrics.
Specifications
NVIDIA DGX B300 specifications
Core specifications below are based on NVIDIA product and DGX system documentation. Facility-specific requirements should be validated against the exact AC or DC configuration quoted.
- Product
- NVIDIA DGX B300
- System type
- 10U enterprise AI server / DGX AI supercomputer
- GPU configuration
- 8× NVIDIA B300 Blackwell Ultra SXM GPUs
- GPU memory
- 8 × 288 GB HBM3e = 2,304 GB installed; NVIDIA lists 2.3 TB total
- FP4 Tensor Core performance
- Up to 144 PFLOPS
- FP8 Tensor Core performance
- Up to 72 PFLOPS
- NVLink
- 2× NVIDIA NVLink Switch systems; 14.4 TB/s aggregate bandwidth
- Processors
- 2× Intel Xeon Platinum 6776P
- System memory
- 2 TB default; up to 4 TB
- Cluster networking
- 8× NVIDIA ConnectX-8 VPI via OSFP; up to 800 Gb/s InfiniBand/Ethernet
- DPU networking
- 2× dual-port NVIDIA BlueField-3 DPU; up to 400 Gb/s InfiniBand/Ethernet
- Boot storage
- 2× ~1.9 TB NVMe M.2
- Internal storage
- 8× 3.84 TB NVMe E1.S self-encrypting drives
- Management
- 1 GbE onboard NIC plus 1 GbE BMC with remote management
- Rack height
- 10U
- Approx. chassis dimensions
- 17.4 in H × 19.0 in W × 35.74 in D (442 × 482.6 × 904.2 mm)
- System weight
- Approx. 370.38 lb (168 kg) AC/PSU version; 271.17 lb (123 kg) busbar version
- Power
- 14.5 kW; AC/PSU version uses 200–240 VAC with 12 power inlets, while the busbar version uses 54 VDC
- Software
- NVIDIA DGX OS, NVIDIA AI Enterprise and NVIDIA Mission Control; enterprise Linux support varies by release
- NVIDIA support
- NVIDIA lists three-year business-standard hardware and software support
Commercial buying guide
NVIDIA DGX B300 price, availability and lead time
For organizations searching for NVIDIA DGX B300 for sale, the purchase decision normally involves more than the server price alone. NVIDIA does not publish a single fixed public list price for DGX B300, so a current commercial quote should account for the exact power configuration, support entitlement, destination, freight, installation and any commissioning services required by the target data center.
What determines DGX B300 cost?
- System configuration: confirm whether the order is for the AC/PDU or DC/busbar version and document all included hardware.
- Support: confirm the NVIDIA enterprise support entitlement, service region and response terms attached to the system.
- Logistics: include insured high-value freight, customs or import requirements where applicable, unloading and final-site delivery.
- Deployment: account for rack preparation, power distribution, network integration, professional installation and commissioning.
- Facility readiness: verify that power, cooling, rack loading and network capacity are available before shipment.
Availability and lead time
DGX B300 availability can change by region, quantity and configuration. A quote should therefore state whether the requested system is available, the estimated dispatch or delivery window, and any conditions that could affect the schedule.
For procurement planning, request current DGX B300 price and availability for the exact quantity and delivery country rather than relying on a historical market price.
Platform comparison
DGX B300 vs DGX B200 and HGX B300
Buyers commonly compare DGX B300 with the previous-generation DGX B200 and with OEM servers built around NVIDIA HGX B300. The right choice depends on memory requirements, networking, deployment model and whether a complete NVIDIA-integrated DGX system or an OEM-integrated HGX platform is preferred.
DGX B300 vs DGX B200
DGX B300 uses Blackwell Ultra B300 GPUs with 288 GB HBM3e per GPU, giving 2.3 TB total GPU memory. DGX B200 uses Blackwell B200 GPUs with 180 GB per GPU, giving 1.44 TB total. Both are eight-GPU DGX systems, but B300 provides substantially more accelerator memory for larger models, longer context and memory-intensive inference or training workloads.
DGX B300 vs HGX B300
DGX B300 is NVIDIA’s complete, branded AI system with validated CPUs, storage, networking, management, DGX software and NVIDIA support. HGX B300 is NVIDIA’s eight-GPU accelerator platform/baseboard used by certified OEMs to build their own B300 servers. HGX can provide more server-vendor and chassis flexibility; DGX provides a standardized NVIDIA-integrated system.
Which one should you buy?
Choose DGX B300 when a standardized NVIDIA platform, DGX software integration and direct DGX lifecycle are priorities. Compare HGX B300 offerings when OEM configuration flexibility, alternative chassis designs or a broader server-vendor procurement strategy matters. Buyers upgrading from DGX B200 should focus especially on the B300’s larger GPU memory capacity and Blackwell Ultra workload profile.
Facility planning
NVIDIA DGX B300 power consumption, cooling and rack requirements
DGX B300 power consumption is a primary procurement consideration. NVIDIA lists system power consumption at approximately 14.5 kW. The AC/PSU configuration uses 200–240 VAC and provides 12 power inlets, while the DC/busbar configuration uses 54 VDC. Facility engineers should validate the exact ordered version before final rack and electrical design.
10U rack space
Reserve 10U and confirm cabinet depth, static and dynamic load limits, service clearance and the installation path for a high-density system.
High-density power
Plan around the quoted AC/PDU or DC/busbar configuration and verify circuit, PDU or busbar capacity with the facility team before delivery.
Cooling and airflow
NVIDIA documents approximately 49,476 BTU/hr heat output at maximum system power and an inlet-air operating range of 10–30 °C. Cooling design should follow NVIDIA data-center guidance for the exact deployment.
Professional installation
NVIDIA states that DGX B300 installation must be performed by NVIDIA partner-network personnel or NVIDIA field service engineers to preserve the hardware warranty.
Ownership decision
Buy DGX B300 or use cloud GPU infrastructure?
Organizations evaluating whether to buy NVIDIA DGX B300 or rent accelerator capacity should compare total workload economics rather than hardware price alone. Owning a DGX system can make sense for sustained, security-sensitive or high-utilization workloads where the organization already has suitable data-center infrastructure and operations expertise. Cloud capacity can be more flexible for experimentation, burst workloads or teams that do not want to operate high-density infrastructure.
DGX ownership may fit when
- GPU utilization is expected to remain high over a long operating period.
- Data locality, privacy, compliance or predictable infrastructure access is important.
- The organization already has suitable rack, power, cooling and networking capacity.
- Internal teams can manage DGX software, networking, monitoring and lifecycle operations.
Cloud capacity may fit when
- The workload is temporary, experimental or highly variable.
- Fast access matters more than owning and installing infrastructure.
- The organization wants to avoid facility upgrades and hardware operations.
- Capacity needs may change quickly and are difficult to forecast.
The correct comparison should include hardware, support, electricity, cooling, facility space, networking, staffing, financing and expected utilization. A procurement team should model those factors against current cloud pricing for its actual workload before choosing either approach.
Deployment checklist
What to verify before ordering NVIDIA DGX B300
A DGX B300 purchase should be treated as a data-center infrastructure project, not as a standard graphics-card installation.
- Power architecture: confirm AC/PDU or DC/busbar version and the data center’s available power capacity.
- Rack capacity: reserve 10U of rack space and verify cabinet depth, load rating and service clearance.
- Cooling and airflow: validate the NVIDIA-recommended thermal design for the selected power/rack configuration.
- Network fabric: plan 800 Gb/s ConnectX-8 cluster networking and 400 Gb/s BlueField-3 connectivity as required by the deployment.
- Software environment: confirm the DGX OS / enterprise Linux release, NVIDIA AI Enterprise licensing and orchestration requirements.
- Installation: NVIDIA states that DGX B300 must be installed by NVIDIA partner-network personnel or NVIDIA field service engineers to preserve the hardware warranty.
- Support: confirm NVIDIA enterprise support entitlement, region, response level and any professional services.
- Delivery: account for high-value freight, insurance, loading/unloading, rack installation and commissioning requirements.
Technical documentation
NVIDIA DGX B300 official resources
Use NVIDIA documentation to validate product specifications and deployment requirements before installation.
DGX B300 product page
Official NVIDIA product overview, specifications, software and platform information.
DGX B300 System User Guide
Official system feature summary, hardware, management and operating information.
Data-center deployment guidance
Official NVIDIA data-center power, rack, dimensions and deployment planning information for DGX B300.
Buyer questions
NVIDIA DGX B300 FAQ
Common questions from organizations comparing DGX B300 price, specifications, availability, power requirements and deployment options.
What is NVIDIA DGX B300?
NVIDIA DGX B300 is a 10U enterprise AI system built around eight NVIDIA B300 Blackwell Ultra SXM GPUs. It combines GPUs, CPUs, system memory, NVLink/NVSwitch, high-speed networking, NVMe storage, remote management and NVIDIA’s DGX software stack in one validated platform.
How much GPU memory does DGX B300 have?
DGX B300 contains eight B300 GPUs with 288 GB HBM3e each. That equals 2,304 GB installed GPU memory, which NVIDIA lists as approximately 2.3 TB total.
How much does NVIDIA DGX B300 cost?
NVIDIA does not publish one fixed public list price for DGX B300. Final DGX B300 cost depends on configuration, support, region, quantity, freight, installation and deployment scope. Request a current destination-specific quotation.
Where can I buy NVIDIA DGX B300?
Organizations looking for NVIDIA DGX B300 for sale should use an authorized or verified enterprise procurement channel capable of documenting configuration, support, logistics and installation requirements. AI Robot Supplier provides B2B quotations subject to current availability and destination.
What is the lead time for NVIDIA DGX B300?
Lead time is not fixed and can vary with region, quantity, power configuration and supply conditions. Confirm estimated dispatch and delivery timing in the written quotation before placing an order.
Is DGX B300 a graphics card?
No. DGX B300 is a complete enterprise AI server and DGX supercomputing system. It is not a standalone PCIe graphics card.
What workloads is DGX B300 intended for?
Typical workloads include generative AI training, large language model inference, reasoning models, fine-tuning, retrieval-augmented generation, multimodal AI, HPC and scale-out enterprise AI services.
How much power does DGX B300 consume?
NVIDIA lists DGX B300 system power consumption at approximately 14.5 kW. Buyers must verify facility power design for the exact AC/PSU or DC/busbar configuration being ordered.
What cooling does DGX B300 require?
DGX B300 is a high-density 10U data-center system. NVIDIA documents approximately 49,476 BTU/hr maximum heat output and an inlet-air range of 10–30 °C. Cooling and airflow should be validated against NVIDIA deployment guidance and the target facility.
What is the difference between DGX B300 and DGX B200?
DGX B300 uses Blackwell Ultra B300 GPUs with 288 GB HBM3e per GPU and approximately 2.3 TB total GPU memory. DGX B200 uses B200 GPUs with 180 GB per GPU and 1.44 TB total. B300 is aimed at newer memory-intensive reasoning, inference and training workloads.
What is the difference between DGX B300 and HGX B300?
DGX B300 is a complete NVIDIA-branded system with validated CPUs, storage, networking, software and DGX support. HGX B300 is NVIDIA’s eight-GPU accelerator platform used by OEM server manufacturers to build their own B300 systems.
Should I buy DGX B300 or use cloud GPUs?
The answer depends on utilization, data-center readiness, security requirements and workload duration. Ownership can suit sustained high-utilization workloads; cloud infrastructure can suit temporary, bursty or experimental workloads. Compare total cost of ownership rather than server purchase price alone.
Can a standard data-center rack support DGX B300?
DGX B300 occupies 10U, but rack suitability also depends on cabinet depth, load rating, power delivery, airflow, service clearance and network design. Validate all facility requirements before ordering.
Does DGX B300 require professional installation?
Yes. NVIDIA states that DGX B300 must be installed by NVIDIA partner-network personnel or NVIDIA field service engineers to preserve the hardware warranty.
What support does NVIDIA list for DGX B300?
NVIDIA lists three-year business-standard hardware and software support for DGX B300. The exact support entitlement, region and service level supplied with an order should be confirmed in the quotation.
NVIDIAManufacturer / brand
Manufacturer information
NVIDIA documentation for DGX B300
NVIDIA develops DGX systems as integrated accelerated-computing platforms for enterprise AI. Product specifications, system software, rack requirements and support terms should be checked against NVIDIA documentation and the exact commercial offer.
AI Robot Supplier
Procurement support for NVIDIA DGX B300
Confirm the exact DGX B300 power version, support scope and supplied hardware before order approval.
Coordinate rack, power, cooling and networking requirements with the target facility before delivery.
Arrange suitable high-value freight, insurance and shipment documentation for approved destinations.
Receive a written commercial quotation defining configuration, delivery scope and agreed terms.
B2B quotation
Request an NVIDIA DGX B300 price and availability quote
Include delivery country, quantity, preferred AC or DC deployment approach, target installation date and any support or commissioning requirements.
