Artificial intelligence is transforming almost every business by automating tasks,
enhancing customer service, generating insights, and enabling innovation. It’s
no longer a futuristic concept but a reality that’s fundamentally reshaping how
businesses operate. However, as AI workloads continue to develop, they’re beginning
to require significantly more compute capacity for both training and inference than
most enterprises have available. To leverage AI, enterprises need high-performance
computing, storage, and networking capabilities that are reliable and efficient.
NVIDIA DGX™ B200 is a complete AI platform that defines the next chapter of
generative AI by taking full advantage of NVIDIA Blackwell GPUs and high-speed
interconnects. Configured with eight NVIDIA Blackwell GPUs, DGX B200 delivers
unparalleled generative AI performance with a massive 1.4 terabytes (TB) of GPU
memory, 64 terabytes per second (TB/s) of HBM3e memory bandwidth, and
14.4 TB/s of all-to-all GPU bandwidth, making it uniquely suited to handle any
enterprise AI workload.
With NVIDIA DGX B200, enterprises can equip their data scientists and developers
with a universal AI supercomputer to accelerate their time to insight and fully realize
the benefits of AI for their businesses.
One Platform for Develop-to-Deploy Pipelines
As AI workflows have become more sophisticated, so too has the need for
enterprises to handle large datasets at all stages of the AI pipeline, from training
to fine-tuning to inference. This requires massive amounts of compute power.
With NVIDIA DGX B200, enterprises can arm their developers with a single, unified
platform built to accelerate their workflows. Supercharged for the next generation
of generative AI, DGX B200 enables businesses to infuse AI into their daily
operations and customer experiences.
Key Features
NVIDIA DGX B200
> Built with eight NVIDIA Blackwell
GPUs
> 1.4 TB of GPU memory space
> 72 petaFLOPS of training
performance
> 144 petaFLOPS of inference
performance
> NVIDIA networking
> Dual 5th generation Intel® Xeon®
Scalable Processors
> Foundation of NVIDIA DGX
BasePOD™ and NVIDIA DGX
SuperPOD™
> Leverages NVIDIA AI Enterprise
and NVIDIA Mission Control
software
enhancing customer service, generating insights, and enabling innovation. It’s
no longer a futuristic concept but a reality that’s fundamentally reshaping how
businesses operate. However, as AI workloads continue to develop, they’re beginning
to require significantly more compute capacity for both training and inference than
most enterprises have available. To leverage AI, enterprises need high-performance
computing, storage, and networking capabilities that are reliable and efficient.
NVIDIA DGX™ B200 is a complete AI platform that defines the next chapter of
generative AI by taking full advantage of NVIDIA Blackwell GPUs and high-speed
interconnects. Configured with eight NVIDIA Blackwell GPUs, DGX B200 delivers
unparalleled generative AI performance with a massive 1.4 terabytes (TB) of GPU
memory, 64 terabytes per second (TB/s) of HBM3e memory bandwidth, and
14.4 TB/s of all-to-all GPU bandwidth, making it uniquely suited to handle any
enterprise AI workload.
With NVIDIA DGX B200, enterprises can equip their data scientists and developers
with a universal AI supercomputer to accelerate their time to insight and fully realize
the benefits of AI for their businesses.
One Platform for Develop-to-Deploy Pipelines
As AI workflows have become more sophisticated, so too has the need for
enterprises to handle large datasets at all stages of the AI pipeline, from training
to fine-tuning to inference. This requires massive amounts of compute power.
With NVIDIA DGX B200, enterprises can arm their developers with a single, unified
platform built to accelerate their workflows. Supercharged for the next generation
of generative AI, DGX B200 enables businesses to infuse AI into their daily
operations and customer experiences.
Key Features
NVIDIA DGX B200
> Built with eight NVIDIA Blackwell
GPUs
> 1.4 TB of GPU memory space
> 72 petaFLOPS of training
performance
> 144 petaFLOPS of inference
performance
> NVIDIA networking
> Dual 5th generation Intel® Xeon®
Scalable Processors
> Foundation of NVIDIA DGX
BasePOD™ and NVIDIA DGX
SuperPOD™
> Leverages NVIDIA AI Enterprise
and NVIDIA Mission Control
software
NVIDIA DGX B200 | Datasheet | 2
Powerhouse of AI Performance
NVIDIA is dedicated to designing the next generation of the world’s most powerful
supercomputers, built to tackle the most complex AI problems that enterprises
face. Powered by the NVIDIA Blackwell architecture’s advancements in computing,
DGX B200 delivers 3x the training performance and 15x the inference performance
of DGX H100. DGX B200 offers high-speed scalability for NVIDIA DGX BasePOD
and NVIDIA DGX SuperPOD, delivering top-of-the-line performance in a turnkey AI
infrastructure solution.
Proven Infrastructure Standard
NVIDIA DGX B200 is the world’s first system with the NVIDIA Blackwell GPU,
delivering breakthrough performance for the world’s most complex AI problems,
such as large language models and natural language processing. DGX B200 offers
a fully optimized hardware and software platform that leverages the complete
NVIDIA AI software stack, a rich ecosystem of third-party support, and access to
expert advice from NVIDIA professional services, allowing organizations to solve the
biggest and most complex business problems with AI.
Run Models, Automate the Essentials With NVIDIA
Mission Control
NVIDIA Mission Control powers every aspect of AI factory operations, from
developer workloads to infrastructure to facilities, with the skills of a world-class
operations team, now delivered as software. It brings instant agility for inference
and training while providing full-stack intelligence for infrastructure resilience.
Mission Control lets every enterprise run AI with hyperscale-grade efficiency,
accelerating AI experimentation. Additionally, NVIDIA AI Enterprise, offering a
suite of software to streamline AI development and deployment, is optimized to
run on NVIDIA DGX systems. Use NVIDIA NIM™ microservices for optimal model
deployment, offering speed, ease of use, manageability, and security.
Seamless workload orchestration
with NVIDIA Run:ai technology
Problem identification, isolation,
and recovery
Customizable dashboards
On-demand health checks
Next-generation cluster management
for AI’s next frontiers
AI Workload Management
Autonomous Recovery Engine
Advanced Cluster Provisioning
Telemetry and Observability
Validation and Diagnostics
NVIDIA Mission Control
NVIDIA AI Enterprise
AI Factory Infrastructure
AI Practitioners
Enterprise IT
State-of-the-art AI factory software stack
0
10
30
40
DGX H100
3.5
58
DGX B200
50
60
70
GPT-MoE-1.8T Real-time Throughput
20
15X
Output Tokens/sec/GPU
Real-Time Large Language
Model Inference
Projected performance subject to change. Token-to-token latency (TTL)
= 50 ms real time, first token latency (FTL) = 5,000ms, input sequence
length = 32,768, output sequence length = 1,028, 8x eight-way DGX
H100 GPUs air-cooled vs. 1x eight-way DGX B200 air-cooled, per GPU
performance comparison.
0X
1X
DGX H100
3X
DGX B200
3X
GPT-MoE-1.8T Model Training Speed-Up
2X
1X
Supercharged AI Training
Performance
Projected performance subject to change. 32,768 GPU scale, 4,096x
eight-way DGX H100 air-cooled cluster: 400G IB network, 4,096x 8-way
DGX B200 air-cooled cluster: 400G IB network.
Powerhouse of AI Performance
NVIDIA is dedicated to designing the next generation of the world’s most powerful
supercomputers, built to tackle the most complex AI problems that enterprises
face. Powered by the NVIDIA Blackwell architecture’s advancements in computing,
DGX B200 delivers 3x the training performance and 15x the inference performance
of DGX H100. DGX B200 offers high-speed scalability for NVIDIA DGX BasePOD
and NVIDIA DGX SuperPOD, delivering top-of-the-line performance in a turnkey AI
infrastructure solution.
Proven Infrastructure Standard
NVIDIA DGX B200 is the world’s first system with the NVIDIA Blackwell GPU,
delivering breakthrough performance for the world’s most complex AI problems,
such as large language models and natural language processing. DGX B200 offers
a fully optimized hardware and software platform that leverages the complete
NVIDIA AI software stack, a rich ecosystem of third-party support, and access to
expert advice from NVIDIA professional services, allowing organizations to solve the
biggest and most complex business problems with AI.
Run Models, Automate the Essentials With NVIDIA
Mission Control
NVIDIA Mission Control powers every aspect of AI factory operations, from
developer workloads to infrastructure to facilities, with the skills of a world-class
operations team, now delivered as software. It brings instant agility for inference
and training while providing full-stack intelligence for infrastructure resilience.
Mission Control lets every enterprise run AI with hyperscale-grade efficiency,
accelerating AI experimentation. Additionally, NVIDIA AI Enterprise, offering a
suite of software to streamline AI development and deployment, is optimized to
run on NVIDIA DGX systems. Use NVIDIA NIM™ microservices for optimal model
deployment, offering speed, ease of use, manageability, and security.
Seamless workload orchestration
with NVIDIA Run:ai technology
Problem identification, isolation,
and recovery
Customizable dashboards
On-demand health checks
Next-generation cluster management
for AI’s next frontiers
AI Workload Management
Autonomous Recovery Engine
Advanced Cluster Provisioning
Telemetry and Observability
Validation and Diagnostics
NVIDIA Mission Control
NVIDIA AI Enterprise
AI Factory Infrastructure
AI Practitioners
Enterprise IT
State-of-the-art AI factory software stack
0
10
30
40
DGX H100
3.5
58
DGX B200
50
60
70
GPT-MoE-1.8T Real-time Throughput
20
15X
Output Tokens/sec/GPU
Real-Time Large Language
Model Inference
Projected performance subject to change. Token-to-token latency (TTL)
= 50 ms real time, first token latency (FTL) = 5,000ms, input sequence
length = 32,768, output sequence length = 1,028, 8x eight-way DGX
H100 GPUs air-cooled vs. 1x eight-way DGX B200 air-cooled, per GPU
performance comparison.
0X
1X
DGX H100
3X
DGX B200
3X
GPT-MoE-1.8T Model Training Speed-Up
2X
1X
Supercharged AI Training
Performance
Projected performance subject to change. 32,768 GPU scale, 4,096x
eight-way DGX H100 air-cooled cluster: 400G IB network, 4,096x 8-way
DGX B200 air-cooled cluster: 400G IB network.
DGX B200 Technical Specifications
GPU 8x NVIDIA Blackwell GPUs
GPU Memory 1,440 GB total, 64 TB/s HBM3e bandwidth
Performance FP4 Tensor Core: 144 | 72* petaFLOPS
FP8 Tensor Core: 72 | 36* petaFLOPS
NVIDIA® NVSwitch™ 2x
NVIDIA NVLink Bandwidth 14.4 TB/s aggregate bandwidth
System Power Usage ~14.3 kW max
CPU 2 Intel® Xeon® Platinum 8570 Processors
112 Cores total, 2.1 GHz (Base),
4 GHz (Max Boost)
System Memory 2 TB, configurable to 4 TB
Networking 4x OSFP ports serving 8x single-port NVIDIA ConnectX-7 VPI
>Up to 400 Gb/s NVIDIA InfiniBand/Ethernet
2x dual-port QSFP112 NVIDIA BlueField-3 DPU
>Up to 400 Gb/s NVIDIA InfiniBand/Ethernet
Management Network 10 Gb/s onboard NIC with RJ45
100 Gb/s dual-port ethernet NIC
Host baseboard management controller (BMC) with RJ45
Storage OS: 2x 1.9 TB NVMe M.2
Internal storage: 8x 3.84 TB NVMe U.2
Software NVIDIA AI Enterprise – Optimized AI Software
NVIDIA Mission Control – AI Data Center Operations and Orchestration With
NVIDIA Run:ai Technology
NVIDIA DGX OS / Ubuntu – Operating System
Rack Units (RU) 10 RU
System Dimensions Height: 17.5 in (444 mm)
Width: 19.0 in (482.2 mm)
Length: 35.3 in (897.1 mm)
Operating Temperature 10-35°C (50-90°F)
Enterprise Support Three-year Enterprise Business-Standard Support for hardware and software
*Specifications shown as sparse | dense.
Ready to Get Started?
To learn more about NVIDIA DGX B200,
visit: nvidia.com/dgx-b200
© 2025 NVIDIA Corporation and affiliates. All rights reserved. NVIDIA, the NVIDIA logo, BlueField, ConnectX, DGX,
DGX BasePOD, DGX SuperPOD, Mission Control, NIM, and NVSwitch are trademarks and/or registered trademarks
of NVIDIA Corporation and affiliates in the U.S. and other countries. Other company and product names may be
GPU 8x NVIDIA Blackwell GPUs
GPU Memory 1,440 GB total, 64 TB/s HBM3e bandwidth
Performance FP4 Tensor Core: 144 | 72* petaFLOPS
FP8 Tensor Core: 72 | 36* petaFLOPS
NVIDIA® NVSwitch™ 2x
NVIDIA NVLink Bandwidth 14.4 TB/s aggregate bandwidth
System Power Usage ~14.3 kW max
CPU 2 Intel® Xeon® Platinum 8570 Processors
112 Cores total, 2.1 GHz (Base),
4 GHz (Max Boost)
System Memory 2 TB, configurable to 4 TB
Networking 4x OSFP ports serving 8x single-port NVIDIA ConnectX-7 VPI
>Up to 400 Gb/s NVIDIA InfiniBand/Ethernet
2x dual-port QSFP112 NVIDIA BlueField-3 DPU
>Up to 400 Gb/s NVIDIA InfiniBand/Ethernet
Management Network 10 Gb/s onboard NIC with RJ45
100 Gb/s dual-port ethernet NIC
Host baseboard management controller (BMC) with RJ45
Storage OS: 2x 1.9 TB NVMe M.2
Internal storage: 8x 3.84 TB NVMe U.2
Software NVIDIA AI Enterprise – Optimized AI Software
NVIDIA Mission Control – AI Data Center Operations and Orchestration With
NVIDIA Run:ai Technology
NVIDIA DGX OS / Ubuntu – Operating System
Rack Units (RU) 10 RU
System Dimensions Height: 17.5 in (444 mm)
Width: 19.0 in (482.2 mm)
Length: 35.3 in (897.1 mm)
Operating Temperature 10-35°C (50-90°F)
Enterprise Support Three-year Enterprise Business-Standard Support for hardware and software
*Specifications shown as sparse | dense.
Ready to Get Started?
To learn more about NVIDIA DGX B200,
visit: nvidia.com/dgx-b200
© 2025 NVIDIA Corporation and affiliates. All rights reserved. NVIDIA, the NVIDIA logo, BlueField, ConnectX, DGX,
DGX BasePOD, DGX SuperPOD, Mission Control, NIM, and NVSwitch are trademarks and/or registered trademarks
of NVIDIA Corporation and affiliates in the U.S. and other countries. Other company and product names may be
https://resources.nvidia.com/en-us-dgx-h100/dgx-b200-datasheet?ncid=em-nurt-978152&mkt_tok=MTU2LU9GTi03NDIAAAGjD8liPljd9fwK8mXRBtEaX_uVGu_qX7l3NmACIWPPOh7Vg3eXCS90aGxiiq_ZP_VxFriq6pLwVi4dl4P4TN9-UMykLBd68fnXUgLKFnMDUxzzmLno1lkZ&_pfses=bd32utMttWS66LMtYCHj78Dw
